Version v2026.10.01 - Daily Briefing

Weather

Bloomington

67°

Clear

H 88°L 65°

Feels like: 64°
Humidity: 55%
Cloud: 9%
Rain: 9%

Today

88°

65°

Fri

77°

56°

67%

Sat

73°

49°

Sun

75°

51°

Mon

67°

46°

Tue

68°

38°

Wed

77°

44°

News

europe/conflict

Ukraine strikes oil facility in Russia's Samara region

Ukraine conducted a strike against an oil facility located in the Samara region of Russia. This maneuver comes despite ongoing geopolitical pressure from the Trump administration for Ukraine to moderate its long-range offensive operations. The attack is part of an intensifying pattern of targeting Russian energy infrastructure, which serves as a financial backbone for their military efforts. The UK has simultaneously announced 31 new sanctions targeting Russian LNG shadow fleet vessels and Kremlin disinformation networks, further tightening the economic net around the Russian state as the conflict shows no signs of cooling down.

Targeting energy infrastructure is a strategic attempt to degrade the economic mechanism fueling the Russian war machine. By disrupting oil processing and logistics, Ukraine aims to force a shift in Moscow’s resource allocation. This approach demonstrates a shift toward asymmetrical warfare, where internal economic stability becomes as vital as battlefield positioning. Such strikes not only cause immediate physical damage but also complicate the global energy supply chain, creating pressure on international markets that rely on Russian output, which complicates the diplomatic landscape for Ukraine’s allies as they attempt to balance economic stability with military support.

Persistent strikes on energy infrastructure force Russia to divert resources away from the front lines to defend its domestic economic assets.

The Guardian

middle east/conflict

Flydubai flight diverted after mid-air violence

A commercial flight operated by flydubai bound for Tel Aviv was forced to divert to Saudi Arabia after a violent altercation broke out in the cockpit. Reports indicate that one pilot allegedly stabbed their co-pilot during the flight, leading to a dramatic emergency landing. While Israeli officials have publicly cautioned against jumping to conclusions regarding the motive or any potential state involvement, the incident has prompted the airline to halt all flights to Israel. The situation remains under investigation as authorities piece together the timeline and the factors that led to such an extreme breach of flight deck security.

The event has triggered significant regional anxiety, highlighting the fragility of aviation security in an already unstable Middle East. If the violence was politically motivated, it represents an expansion of the conflict into civilian airspace, a scenario that historically leads to drastic changes in international travel regulations and security protocols. Regardless of the pilot's intent, the incident forces carriers to re-evaluate their vetting and monitoring processes for personnel operating in sensitive regions. The immediate suspension of flights acts as a chilling effect on regional connectivity, isolating trade and diplomatic movement between Israel and its neighbors.

This incident forces an urgent reassessment of cockpit security and aviation safety protocols across the entire Middle East region.

BBC NewsThe Guardian

asia/economy

South Korea exports reach record highs on AI demand

South Korea’s export economy has hit a record high, surpassing 120 billion dollars in value for the first time. This surge is almost entirely driven by the global appetite for artificial intelligence technology and the resulting, ferocious demand for semiconductors. As AI models become more complex, the hardware required to train and run them has become the most valuable commodity in the tech supply chain. South Korean manufacturers have positioned themselves at the heart of this boom, supplying the essential processing power that global tech giants currently scramble to acquire at any price.

This export milestone highlights how deeply national economies are now tied to the development of specific emerging technologies. By focusing on high-end hardware, South Korea is effectively acting as the engine room for the global AI revolution. While this generates massive short-term revenue, it also leaves the domestic economy highly vulnerable to fluctuations in the tech sector. If the momentum behind AI investment slows or if geopolitical trade barriers shift, the current rapid growth could face a sudden contraction, forcing a painful pivot for the country's manufacturing-heavy industrial strategy.

The concentration of high-end manufacturing in one sector makes the South Korean economy uniquely sensitive to the volatility of global tech spending.

Al Jazeera

europe/climate

Swiss glaciers lose 5 percent of ice in single year

Scientific data shows that Switzerland's glaciers have lost five percent of their total ice volume over the past year. This loss is attributed to a series of intense heatwaves that have pushed summer temperatures to record-breaking levels. Researchers describe the current rate of melting as catastrophic, warning that the pace is far exceeding previous climate models. As these glaciers shrink, they leave behind barren landscapes and remove a critical natural reservoir for freshwater that traditionally supplies the surrounding regions throughout the drier months of the year, fundamentally changing the alpine topography.

The mechanism of glacier melt is a self-reinforcing process; as ice sheets retreat, they reveal darker surfaces that absorb more solar radiation, which then accelerates further melting. This is not just a loss of scenery; it is the physical disappearance of a long-term water storage system. For downstream agriculture and hydroelectric power, this loss means an increasingly unreliable water supply. As the glaciers vanish, the ability to manage water shortages during future heatwaves will become significantly more difficult, forcing policy makers to confront the permanent loss of a foundational environmental buffer.

Accelerated glacial retreat disrupts water security for downstream regions, permanently altering local agriculture and energy generation capabilities.

Al Jazeera

americas/technology

OpenAI reports failed hacking attempt on Canadian government

OpenAI is currently reviewing reports of a failed cyberattack directed at the national archives agency of Canada. The incident involved an attempt by AI-powered bots to gain unauthorized access to government databases. While the company stated that the attempt was unsuccessful, the event underscores the growing vulnerability of government data to automated exploitation tools. OpenAI is investigating how its models were utilized in the attempt and what, if any, security protocols were bypassed by the actors behind the attack. The Canadian government is monitoring the situation to ensure no sensitive historical or administrative data was compromised.

The core issue here is the democratization of sophisticated hacking tools via artificial intelligence. By leveraging machine learning, attackers can perform reconnaissance and vulnerability scanning at a scale and speed that human operators cannot match. This forces government agencies to fundamentally rethink their defense architecture, moving away from static perimeter security toward dynamic systems that can identify and block AI-driven anomalous behavior in real-time. This is no longer just a technical challenge; it is a race to build security infrastructure capable of outpacing the automated threats it aims to neutralize.

The use of automated AI agents to probe government infrastructure creates a new, faster-moving threat environment that current cyber defenses are not optimized to address.

Al Jazeera

asia/conflict

Pakistan and Afghanistan trade deadly cross-border air strikes

Pakistan has launched air strikes into Afghanistan, reporting the deaths of 22 fighters it describes as militants. However, officials in Kabul have offered a different account, stating that the strikes in Kunar and Helmand provinces resulted in the deaths of nine civilians, including women and children. This flare-up highlights the increasingly tense border security situation between the two countries. The strikes follow recurring disputes over the presence of armed groups operating near the frontier, with both sides accusing the other of either harbor or state-sponsored aggression, further fraying diplomatic ties that were already fragile.

The mechanism of this conflict is rooted in the lack of a clear, mutually respected border and the divergent goals of security forces on each side. When one country conducts unilateral air strikes, it erodes the sovereignty of the other, leading to a cycle of retaliatory violence that makes regional stability impossible. For the local populations living in these provinces, the military activity creates a constant, low-level humanitarian crisis. As long as there is no formal mechanism for de-escalation, these strikes serve as a proxy for larger national security anxieties, often resulting in tragedy for civilians.

Unchecked cross-border military operations threaten to escalate into a broader regional conflict, further isolating the affected border populations.

Al Jazeera

europe/climate

Swiss glaciers lose 5 percent of their ice volume in a single year

Swiss glaciers have recorded another disastrous year of ice loss, with more than five percent of their total volume disappearing in the last twelve months. Scientists note that this rate of decline has accelerated significantly, making such extreme melting patterns appear increasingly consistent with current climate trends. The loss is not just a visual change in the landscape but a functional crisis for the region.

Glaciers serve as critical water reservoirs for many European regions, slowly releasing meltwater throughout the warmer months to sustain ecosystems and agriculture. When these glaciers shrink this rapidly, they can no longer provide the same reliable supply, potentially leading to long-term shortages. This cycle of melting creates a dangerous vulnerability for water security across the continent as these frozen resources continue to dwindle at an unsustainable pace.

The rapid loss of these reservoirs threatens the long-term stability of freshwater supplies for millions living in downstream regions.

BBC ScienceNPR

middle east/conflict

Pilot stabs co-pilot on Flydubai flight forcing emergency landing

An Israel-bound Flydubai flight was forced to make an emergency landing in Saudi Arabia after a pilot allegedly attacked his co-pilot mid-air. Passengers reported a terrifying experience, describing a sudden drop in altitude and the realization that a violent confrontation had occurred in the cockpit. Israeli officials are now investigating whether the incident, which involved a pilot stabbing a colleague, was a deliberate act of terrorism.

The aviation industry relies on absolute trust and coordination between flight crew members to ensure the safety of hundreds of people at once. When that hierarchy is subverted by violence, the standard safety protocols become nearly impossible to execute effectively. This event has prompted high-level concern from government leaders regarding security vulnerabilities during international commercial travel, particularly given the geopolitical tensions surrounding flights between these specific nations.

This incident exposes critical vulnerabilities in cockpit security protocols during sensitive international flights.

The GuardianBBC News

americas/politics

California moves to ban child marriage in latest legislative shift

California has officially banned child marriage, moving to close a legal loophole that had previously permitted the practice. Governor Gavin Newsom signed the measure into law, framing it as a necessary step to protect minors from exploitation and ensure their autonomy. This decision brings the state in line with a growing movement to eliminate child marriage across the United States, where it remains legal in over thirty other states.

Child marriage laws often vary by jurisdiction due to historical interpretations of parental consent and legal tradition. By standardizing this prohibition, California is challenging the legal norms that allow minors to enter into adult contracts that carry lifelong implications. This policy shift forces a national conversation about the definitions of consent and legal maturity, moving toward a more uniform standard that prioritizes the protection of children over historical precedents that permitted these unions.

This legal change increases pressure on remaining states to reconcile their marriage statutes with modern standards of minor protection.

BBC News

americas/health

Measles outbreak in Pennsylvania leads to fifth confirmed death

Pennsylvania health authorities have confirmed a fifth death associated with a growing measles outbreak that has persisted for several weeks. Cases of the highly contagious virus have more than doubled since August, sparking significant friction between local health agencies and federal officials. Disagreements over the official death count and the management of the outbreak have complicated the public response as infection numbers continue to rise across the state.

Measles is one of the most contagious viral diseases known, spreading rapidly through respiratory droplets. When vaccination rates fall below specific thresholds, communities lose herd immunity, allowing the virus to establish a foothold. The struggle to track and manage this outbreak highlights the difficulty of maintaining public health compliance in an era of growing institutional distrust. These fatalities underscore the severity of the illness, which had been largely controlled in the past.

The death toll highlights the risks of declining vaccination rates and the challenge of managing infectious disease outbreaks during periods of institutional skepticism.

Al Jazeera

africa/conflict

Escalating conflict in Ethiopia leaves 52 civilians dead in Tigray

Fighting in Ethiopia’s Tigray region has escalated, resulting in the deaths of at least 52 civilians, according to medical workers on the ground. The conflict, which has been ongoing for some time, has recently bled into neighboring territories, creating instability and stranding travelers in popular areas like Lalibela. The violence has caused immense suffering for the local population and has significantly disrupted access to basic humanitarian resources in the affected zones.

Regional conflicts often rely on complex networks of local militias and fractured national alliances, making them difficult to resolve through standard diplomatic channels. When fighting spreads to tourist-sensitive areas, it signals a deeper breakdown in regional control and security. This escalation risks deepening the cycle of displacement and famine, which have already characterized the conflict for years. The lack of clear communication corridors makes it harder to provide relief to civilians trapped in the middle of these hostilities.

The geographical expansion of the violence suggests that efforts to contain the war have failed, threatening the broader stability of the Horn of Africa.

BBC News

europe/technology

UK universities warned by MI5 over Chinese state espionage

The British security service, MI5, has issued a formal warning to UK universities regarding risks posed by a Chinese state-owned conglomerate. The alert suggests that research facilities are being targeted for espionage, specifically to steal sensitive data and technological breakthroughs. Universities, many of which have become increasingly reliant on international funding streams, are now facing difficult choices as they balance their financial needs against significant national security concerns posed by these partnerships.

Academic institutions are often hubs of open innovation, but that transparency makes them attractive targets for actors seeking to gain intellectual property. When research is conducted alongside state-backed partners, the line between legitimate collaboration and state-level data acquisition often becomes blurred. This alert forces universities to reconsider their funding models and tighten security, potentially slowing down the pace of collaborative research while safeguarding sensitive British intellectual assets from foreign acquisition.

The crackdown threatens to disrupt established funding flows in higher education while heightening barriers to international scientific cooperation.

The Guardian

World Clocks

San Francisco

—

New York

—

London

—

Berlin

—

Dubai

—

Mumbai

—

Shanghai

—

Sydney

—

Markets

Markets closed on a mixed note, with tech largely buoyed by the release of Google's latest Gemini model and ongoing AI infrastructure momentum. While tech shares held steady, the broader market faced pressure from geopolitical anxiety and cooling sentiment toward legacy benchmarks like the Dow.

META: The company is currently defending itself against allegations that its Muse AI agent accessed private user messages without authorization.

Large-cap technology

S&P 500 by sector

The Debate

Is the pursuit of universal norms in social organization fundamentally a homogenizing force that erodes valuable local cultural variation?

The answer determines whether global coordination efforts inherently require sacrificing the unique cultural frameworks that sustain human pluralism.

Universalism protects diversity

Universal norms actually provide the bedrock of security and human rights necessary for individuals to explore and maintain their specific cultural identities without fear. By establishing a baseline of dignity and freedom, we prevent the oppression that often occurs within insular or authoritarian local cultures. Once basic needs and rights are guaranteed across borders, communities are freed from survival-level pressures and can focus on authentic cultural flourishing. A global framework does not demand uniformity in values; rather, it creates a stable, peaceful environment that allows diverse traditions to persist and interact without the threat of being extinguished by larger, more dominant neighboring cultures or internal despotic actors.

Uniformity erodes culture

Universal norms act as a solvent, breaking down the specific customs and social hierarchies that provide meaning and cohesion to distinct societies. When global institutions impose standardized definitions of rights or success, they inevitably marginalize traditional knowledge and practices that do not fit the technocratic, modernist template. This push for coherence demands that all societies justify themselves in the language of the dominant global order, forcing them to shed their unique, context-dependent social structures. Consequently, the attempt to create a singular human order ignores the wisdom of centuries-old localized adaptations, replacing them with a thin, artificial consensus that lacks deep roots in any actual lived experience or ancestral history.

cruxWhether a universal moral baseline acts as a neutral platform for diversity or as a coercive instrument of cultural convergence.

LessWrong

Should regulatory bodies mandate full transparency for the internal reasoning processes of artificial intelligence systems deployed in high-stakes public services?

This choice pits the need for democratic accountability and safety against the potential for gaming the system and stifling innovation.

Mandate full transparency

When AI determines outcomes for veterans' benefits, deportations, or criminal justice, the public has a right to know the logic behind those decisions. Closed-box models prevent effective oversight and allow algorithmic bias to hide behind a facade of mathematical objectivity. By mandating full transparency, we ensure that these systems are subject to the same standards of legal review as human bureaucrats. If a model cannot be explained, it should not be empowered to make decisions that fundamentally alter lives. Transparency acts as a necessary check on power, ensuring that errors in reasoning can be identified and that citizens have a pathway to challenge automated decisions they believe are unfair or inaccurate.

Prioritize model integrity

Mandating full transparency for complex models often forces a choice between usable safety and vulnerability to manipulation. If a system is forced to be perfectly legible to human evaluators, it may become susceptible to 'evaluation-gaming' or poisoning, where actors exploit the model's transparency to force desired, yet dangerous, outcomes. Furthermore, true technical transparency for advanced models is often physically impossible due to the sheer scale of modern parameters. Imposing unrealistic transparency requirements may drive development underground or towards closed-weight models that ignore safety protocols. Instead of forcing legibility, we should focus on robust, out-of-distribution testing and third-party monitoring that evaluates the behavior of the model as a complete, functioning system without demanding total internal exposure.

cruxWhether the risks of adversarial exploitation of transparent models outweigh the democratic necessity of explaining automated state power.

LessWrongLessWrong

Does the persistent memory of an individual constitute their identity, or is identity fundamentally tied to the continuity of a physical organism?

The answer shifts the understanding of what it means to 'survive' a radical transition, impacting everything from medical ethics to future potential digital transfers.

Identity is memory

Identity is a psychological construct rather than a biological one. If we imagine a future technology where a mind is perfectly replicated or transferred, the person who experiences those memories and maintains that personality is the same individual as the original. Physical continuity is mere biological happenstance; what matters is the continuity of consciousness, values, and the narrative thread of an individual's life. If a person were to lose all memory and personality due to trauma, we would rightly say the original person is gone, even if the body lives on. Thus, the persistent pattern of information and memory is the only meaningful locus of who someone truly is.

Identity is biological

Identity is inseparable from the material, biological history of the individual organism. We are living creatures defined by our metabolism, our nervous systems, and our unique position in the physical world. A digital scan of memories may capture information, but it fails to capture the 'self' because it removes the entity from the causal stream of physical existence. The feeling of being a continuous person depends on the uninterrupted functioning of a specific body that has evolved and aged over time. To ignore the physical substrate is to mistake a map for the territory. Without the physical organism, you have a copy of a person, but not the person themselves.

cruxWhether a person is an informational pattern that can be instantiated elsewhere or an inseparable feature of a specific biological history.

From the Labs

Google DeepMind/lab

Google announces Gemini 4 Argon, a new frontier model for complex workflows

Google DeepMind has introduced Gemini 4 Argon, a frontier AI model optimized for deep reasoning across long-horizon tasks such as cybersecurity, enterprise finance, and legal research. The model significantly increases its output token limit to 1 million tokens, up from 64,000, allowing it to sustain complex, multi-step problem solving. In testing, Argon achieved a 77.9% score on the DeepSWE v1.1 benchmark for software engineering and a 91.7% score on the LVBench for long video understanding. To aid cybersecurity, Google is providing select partners in its Fairwind Program access to the model without standard cyber guardrails, enabling autonomous vulnerability discovery and remediation. The model is currently priced at $2 per million input tokens and $10 per million output tokens.

Argon represents a shift toward models capable of maintaining coherence over massive datasets and long-duration reasoning, which is essential for tasks like auditing entire codebases or analyzing long-form legal documents. By allowing the model to generate huge amounts of text in a single trajectory, Google aims to solve problems that previously required human intervention or complex agentic chains. The decision to release an uncensored version of the model to cyber defenders underscores a high-stakes approach to security, betting that the utility of autonomous patching outweighs the risks. While Google highlights Argon’s lead over competitors like OpenAI’s Astra and Anthropic’s Fable in benchmarks, it remains to be seen how these performance gains will translate into broader enterprise adoption and real-world safety at scale.

Google DeepMind

Flow Engineering/funding

Flow Engineering raises $50M at $750M valuation for AI hardware design

Three-year-old San Francisco startup Flow Engineering secured a $50 million Series B round, pushing its valuation to $750 million. The round was co-led by Antonio Gracias of Valar Equity Partners and Gavin Baker of Atreides Management, with continued participation from Sequoia Capital and individual investor Roelof Botha. The company specializes in AI agents designed to automate the alignment of complex hardware CAD drawings with product requirements and simulation results. Current clients leveraging this technology include major aerospace and automotive entities such as Anduril, Rivian, Joby Aviation, General Motors PPU, and the RV Tech joint venture.

This funding highlights the growing interest in applying agentic AI to the physical world, specifically the often-bottlenecked hardware design cycle. By integrating AI directly into the engineering workflow, Flow aims to reduce the friction between initial design specs and final testing outcomes. While the backing from prominent investors signals strong confidence, the long-term success of these agents in replacing traditional, high-stakes engineering precision remains to be seen. The company now occupies a significant position in the competitive landscape of industrial AI, though it must prove that its automated outputs can consistently meet the rigorous safety and performance standards required by its blue-chip customer base.

TechCrunch

Hugging Face/lab

New Open TTS Leaderboard uses objective metrics to rank voice models

Hugging Face has launched the Open TTS Leaderboard to standardize the evaluation of the 8,000-plus text-to-speech models currently on its hub. While human preference arenas like MOS are the gold standard, they scale slowly and are often biased toward commercial API providers. The new leaderboard shifts to objective metrics, including Word Error Rate (WER) for English intelligibility, Character Error Rate (CER) for character-based languages, and speaker similarity (SIM) scores for voice cloning. Models like hexgrad/Kokoro-82M, Supertone/supertonic-3, and fishaudio/s2-pro currently lead in English performance, while k2-fsa/OmniVoice and others excel in multilingual tests.

This move provides a much faster, scalable way to compare model performance, reducing evaluation cycles from weeks to just hours. By offering a 'Listen' tab, the platform bridges the gap between raw metrics and human perception, allowing the community to vet outputs directly. Although objective scores do not perfectly capture naturalness or expressiveness, they establish a critical, repeatable baseline for a fragmented open-source ecosystem. The leaderboard serves as a tool for developers to balance speed, size, and audio quality, ensuring that open-weights models are more fairly represented against closed-source alternatives. It remains a work in progress that relies on community feedback to ensure evaluations stay meaningful.

Hugging Face

Neko Health/startup

Spotify founder's Neko Health expands into US market after $700M raise

Neko Health, founded by Spotify creator Daniel Ek, has officially brought its preventative body-scanning technology to the United States after securing $700 million in funding. The company is part of a growing wave of consumer-health startups attempting to shift the medical focus toward early detection. Other notable players currently entering this space include Midjourney, which is developing its own proprietary scanning hardware, and Function Health, which has also raised substantial capital to scale its preventative health platform. These companies are betting that data-driven, proactive scanning will eventually become a standard feature of personal wellness.

The primary question remains whether these consumer-led initiatives can successfully integrate into the broader, established healthcare system. Investors like Farooq Abbasi of Preface Ventures argue that these startups represent a fundamental change in how we approach disease, though the long-term utility of the scans is still being evaluated by the industry. The challenge for Neko Health and its competitors is to move beyond the novelty of their technology and prove their clinical efficacy. While the funding numbers are massive, it is still unclear how regulators and doctors will view these external diagnostics when it comes to formal patient care and insurance coverage.

TechCrunch

Google Research/lab

Google introduces Diffusion Controller for precise AI image generation

Google Research has released Diffusion Controller, an add-on network designed to act as a steering damper for text-to-image AI models like Nano Banana, Stable Diffusion, and Flux. Current steering methods are often disconnected, forcing developers to choose between inference-time techniques that may lack precision and heavy fine-tuning methods like LoRA or policy gradients. Diffusion Controller fixes this by treating the denoising process as a continuous control problem. By attaching this lightweight network to a frozen, pre-trained model, engineers can inject precise steering corrections during image generation. In tests, the fully unlocked version of this framework achieved a 90% win rate against baseline models when assessed on human preference benchmarks.

This tool matters because it allows for high-precision control over closed-source or black-box models without requiring users to retrain the underlying engine. By keeping the main model frozen, it preserves the original image quality and visual stability while ensuring the output aligns with specific user prompts. For example, it can force a model to generate a lizard wearing sunglasses without distorting the creature's facial structure or natural proportions. While this marks a significant step forward in unifying model optimization, the effectiveness depends heavily on the quality of the steering instructions provided. This provides a more principled, mathematical language for developers who are currently relying on trial-and-error to balance image quality with intent.

Google Research

NVIDIA/lab

NVIDIA Kumo Tabular releases as an open foundation model for tabular data

NVIDIA Kumo Tabular is a new open foundation model designed for tabular classification and regression that operates in a single forward pass without requiring training, tuning, or manual feature engineering. It utilizes a Transformer architecture with column, row, and in-context attention to process tables, featuring a specialized cell embedding method that uses Fourier features for numerical and categorical values. The model comes in three sizes, ranging from 28 million to 215 million parameters, and is released under the OpenMDW-1.1 license. It currently holds the top ranking on four major industry benchmarks: TabArena, BeyondArena, TALENT, and ScoringBench.

This release addresses the limitations of gradient-boosted trees, which have dominated tabular tasks for two decades but require labor-intensive, task-specific pipelines. By using in-context learning—where the model learns from examples provided in the prompt rather than weight updates—Kumo Tabular allows for a more flexible, generalized approach to enterprise data like transaction logs or sensor data. While the model handles missing values natively and scales to larger tables through length-aware attention, it remains to be seen how it performs against highly specialized, hand-tuned models in unique, non-artificial production environments. The ability to generate point predictions and uncertainty estimates from a single forward pass marks a significant shift in how machine learning teams might manage tabular workflows.

Hugging Face

Multiverse Computing/lab

ProvenanceGuard improves factuality for agents using the Model Context Protocol

ProvenanceGuard is a new verification layer designed to prevent cross-source conflation in AI agents that use the Model Context Protocol (MCP). In modern agentic systems, models pool data from various sources—such as patient records or medical journals—into a single context, which often leads to correct facts being attributed to the wrong source. ProvenanceGuard operates after a model generates an answer by breaking it into specific claims, verifying which source supports each claim, and then ensuring the internal attribution matches the source actually used. Evaluated on 281 real traces from a medical agent, the system employs components like MiniLM for source relevance and DeBERTa for NLI verification to ensure claims are grounded in the correct tool output.

By enforcing strict source-awareness, ProvenanceGuard addresses a critical failure mode where technically accurate information becomes dangerous due to incorrect provenance. In sensitive settings like healthcare, misattributing a clinical observation from a general research paper to a patient's specific history could lead to flawed medical decisions. While the current evaluation used a local configuration to ensure a conservative, high-accuracy decision policy, the method is designed to be adaptable for teams using hosted cloud models. Although this approach adds a layer of post-generation processing, it provides a necessary check against the tendency of black-box agents to hallucinate the origins of the information they present to users.

Hugging Face

DoorDash/startup

DoorDash unveils drone delivery service DoorDash Air

DoorDash has introduced its new drone delivery service, DoorDash Air, utilizing a custom six-propeller aircraft designed for autonomous delivery. The company prioritized building ground-level infrastructure—such as kitchen hand-off processes and packaging systems—before developing the drone hardware itself. To manage logistics, the team leveraged their existing Autonomous Delivery Platform, which uses smart routing to factor in weather, flight range, and ground traffic when determining if an order is suitable for aerial transport. The system was specifically sized based on 13 years of historical order data covering weight and travel distance. Early pilot programs are scheduled to launch in Northern California with partners including Popeyes Louisiana Kitchen, Momo N Curry, and Chipotle.

This shift to aerial delivery is a calculated move to optimize efficiency for smaller, shorter-distance orders that are already integrated into the company’s vast delivery marketplace of 500,000 restaurants. By focusing on systemic logistics rather than just the hardware, DoorDash aims to bridge the gap between human couriers and sidewalk robots. While the company has secured Part 135 air carrier certification from the FAA to operate legally, the ultimate success of the service remains tethered to its ability to reliably handle real-world complexities like high-density urban environments. Whether these drones can provide a consistent and cost-effective alternative to traditional delivery methods will only be clear once the Northern California pilot results begin to roll in.

TechCrunch

Google/lab

Google releases Gemini 4 Argon, claims top spot across AI benchmarks

Alphabet’s latest model, Gemini 4 Argon, is now rolling out to partners through the Fairwind Program. Google claims the model outperforms current industry standards, specifically citing leads over OpenAI’s GPT-6 Astra and Anthropic’s Fable and Opus models on the Vals Index. This index weights performance across finance, coding, legal, and tax sectors relative to their U.S. GDP contributions. Internally, Google staff are already using the model to assist with code debugging and large-scale migrations. Beyond text, Argon demonstrates strength in visual analysis, including chart interpretation and multi-document synthesis. The release follows a period of intense competition between major labs, with Google positioning Argon as its most capable tool yet for both developer productivity and defensive cybersecurity operations.

The competitive landscape of AI is moving rapidly, with major labs releasing new flagship models every few months. By leveraging benchmarks like the Vals Index, Google is trying to frame its progress in terms of economic impact rather than just technical capability. The move also signals a maturing of the AI race, where companies are increasingly focused on enterprise-grade utility—coding, tax, and law—rather than just conversational prowess. It is notable that while Google emphasizes the model's defensive cyber capabilities, the tech industry continues to grapple with the tension between accelerating autonomous AI capabilities and the potential for these systems to be misused. Whether these models actually provide a decisive advantage in enterprise environments remains the true test for Google's latest investment.

TechCrunch

The Pentagon/essay

Pentagon launches Project Meridian to study future of autonomous warfare

Defense Secretary Pete Hegseth has launched Project Meridian, a 120-day initiative tasked with identifying how technological advancements will shape future battlefields and identifying actionable solutions for U.S. military dominance. The project is co-led by Elon Musk, Anduril founder Palmer Luckey, and former House Speaker Newt Gingrich, working under the guidance of Defense Department CTO Emil Michael. Concurrently, the Pentagon announced the formation of an Autonomous Warfare Command intended to speed up the deployment of autonomous systems and drones. While the project is intended to shape future military procurement and development strategies, the Pentagon has provided limited information regarding the partner organization facilitating the project or the specific number of participants involved in the advisory team.

The appointment of Musk and Luckey highlights the administration's pivot toward Silicon Valley-style defense tech, relying on entrepreneurs who are already deeply integrated into the military-industrial complex. While this partnership aims to shorten the gap between private sector innovation and battlefield application, it raises questions about potential conflicts of interest, as both men lead companies that stand to benefit from the very procurement strategies they are helping to define. The project seeks to move quickly, but the reliance on private sector interests to define state-level military policy is a significant change in how the Department of Defense operates. It remains unclear how these findings will be implemented or if they will face resistance from established defense contractors or congressional oversight.

TechCrunch

Factory/startup

CEO of coding startup Factory accuses board adviser of spying for rival

Matan Grinberg, CEO of Factory, publicly fired board adviser Chris Degnan, alleging that Degnan shared confidential product and roadmap information with a direct competitor, Cognition. Degnan, a former Snowflake executive and partner at RPT Partners, subsequently announced he joined Cognition as its chief revenue officer. Degnan denies the spying allegations, claiming he resigned from his advisory role before accepting the new position and that he never leaked sensitive data. The conflict escalated on social media, involving prominent investors like Vinod Khosla, whose firm, Khosla Ventures, maintains financial stakes in both companies.

The public dispute exposes the fragility of governance in the high-stakes AI sector, where venture firms often back competing companies simultaneously. While the legal and ethical nuances of the situation remain contested, the incident highlights the difficulty of maintaining strict firewalls when key personnel move between rivals. It complicates the broader VC landscape, where the traditional model of singular oversight is being tested by the rapid proliferation of AI startups. Ultimately, the industry is left wondering how much internal strategy is really protected when investors and board members are positioned across multiple, often overlapping, interests in the same red-hot market.

TechCrunch

Neko Health/funding

Investor perspective on Neko Health's $700M bet on preventative scanning

Neko Health's $700 million funding round has drawn significant attention from venture capitalists interested in the future of preventative healthcare. Farooq Abbasi, an investor in the company and partner at Preface Ventures, suggests that Neko is leading a charge to make full-body diagnostics a regular consumer experience. The company sits alongside firms like Function Health and Midjourney in a market race to define how data-driven health monitoring should look. The conversation around this investment centers on how Neko’s unique scanning technology performs against other consumer health startups and what specific milestones are required for these diagnostic tools to be accepted by the traditional medical community at large.

The shift toward consumer-facing body scans is significant because it attempts to offload diagnostic work from hospitals to high-tech, private startups. Investors are clearly banking on the idea that users want more control over their own physiological data, provided it is accessible and actionable. However, there is still a massive gap between the current state of these services and full integration into formal healthcare systems. The fundamental uncertainty is whether the data generated by these scans will actually lead to improved health outcomes or simply generate noise for physicians. The success of this $700 million bet will rely on Neko’s ability to prove that its scanning technology can provide clinical value that justifies such a significant valuation.

TechCrunch Venture

Defense Manpower Data Center/lab

Millions of US military records stolen in long-term data breach

The U.S. Defense Manpower Data Center (DMDC) recently suffered a months-long data breach that exposed the unencrypted personal information of approximately 2.8 million living individuals and nearly 300,000 deceased persons. Unauthorized users exploited a security vulnerability within a file-sharing system between October 2025 and mid-July 2026. The stolen data includes Social Security numbers, names, dates of birth, sex, race, and sensitive details regarding military service. The DMDC is responsible for maintaining records for over 60 million current and former service members, civilian staff, and their families, functioning as the primary identity management provider for Department of Defense systems, including building access and smart card credentials.

This incident follows a series of high-profile breaches affecting federal workers, including a recent theft at the FBI. While the Department of Defense has not confirmed any evidence of data misuse, the exposure of such deep-level identity information presents a significant counterintelligence risk. History has shown these datasets are often targeted by foreign actors to profile or coerce government employees, similar to the 2015 Office of Personnel Management breach that affected 22 million people. The breach highlights the persistent fragility of federal identity systems, as the Pentagon has yet to identify the hackers or explain how they determined the data had not been misused.

TechCrunch

BMW/startup

BMW's new electric i3 is priced lower than its gas counterpart

BMW has reached a significant parity milestone with the 2027 3 Series, offering both gas and electric versions of the vehicle. The electric i3 50 xDrive is priced at $61,500, while its gas-powered counterpart, the M350 xDrive, costs $65,900. This makes the EV model 6.7% cheaper than the gas version by a margin of $4,400. While the gas-powered car uses an overhauled existing platform, the i3 benefits from BMW’s newest chassis and technology. Both models offer similar power and styling. The i3 provides an estimated 468 miles of range, with fast-charging capabilities that can add 208 miles of range in roughly 10 minutes, making it highly competitive with the 450-mile range of the gas model.

For years, analysts predicted that electric vehicles would reach price parity with gas cars by the mid-to-late 2020s. BMW’s pricing strategy serves as a practical proof point for this shift, moving the conversation from total cost of ownership to immediate, front-end purchase price. By providing two nearly identical cars on different platforms, the company has stripped away much of the guesswork for consumers comparing fuel sources. While purists may still prefer the refueling speed of internal combustion, the gap is narrowing rapidly. It remains to be seen if other automakers can replicate this pricing structure across their broader fleets, especially as battery technology continues to evolve and supply chains become more efficient for EV production.

TechCrunch

ElevenLabs/startup

ElevenLabs valuation reaches $22 billion

ElevenLabs just hit a $22 billion valuation, marking a sharp jump from the $11 billion figure it reached during a $500 million raise in February. This new valuation emerged through a $300 million tender offer co-led by Wellington and T. Rowe Price, which allowed current employees to cash out some of their vested equity. This is the fourth year the company has been active, and it is already their second time running a secondary transaction. They previously held a $100 million tender in September 2025, which at the time placed the company's valuation at $6.6 billion. The firm maintains offices in both New York and London and focuses on high-fidelity synthetic voice and sound effect generation.

This secondary offering is less about raising capital for the company and more about keeping talent locked in. By letting employees monetize their stakes now, ElevenLabs is using liquidity as a retention hook to stop people from jumping to rival AI firms. It is a playbook we are seeing across the board for high-growth tech outfits trying to keep their teams stable before an eventual public offering. While Wellington and T. Rowe Price clearly see a path to a long-term win by holding these shares, the broader question is whether the market can keep supporting these massive valuations as the novelty of synthetic voice tech settles into a competitive, crowded utility market.

TechCrunch

OpenAI/lab

OpenAI introduces Decisions API to speed up agentic tasks

OpenAI announced a new Decisions API that mimics the functionality of Jev, a specialized model from TypeSafe AI. The API allows developers to provide the Luna model with a predefined set of choices, enabling it to classify images or execute agent behaviors at higher speeds and lower costs. By focusing the model on a specific set of outcomes rather than open-ended generation, OpenAI aims to improve performance for automation tasks. Cybersecurity expert Shapor Naghibzadeh demonstrated that this type of focused decision-making could be used to monitor AI agents for malicious actions. In a hackathon demo, he showed that using Jev for security monitoring cost $2.94, compared to $372 using a standard frontier large language model.

The shift toward these specialized decision models represents a push to solve the high cost and latency issues currently plaguing complex AI agents. While current LLMs are powerful, they are often too slow and expensive to monitor every single action an agent takes on the internet. By offloading these tasks to smaller, highly optimized models, labs like OpenAI can introduce a layer of security that is economically viable to run in real-time. The main challenge moving forward will be ensuring these models are accurately calibrated to distinguish between safe and dangerous behaviors without losing the intelligence that makes the larger systems useful. As competition grows, the ability to balance speed, safety, and cost will define the next generation of AI infrastructure.

TechCrunch

Hacker News

1384 points891 comments

Gemini 4 Argon

Google has introduced Gemini 4 Argon, their new frontier model designed for complex, long-horizon workflows in software engineering, legal, finance, and cybersecurity. A major technical change is the expansion of the output token limit to 1 million tokens, a significant leap from the previous 64K limit. This increase is intended to provide the model with more headroom for deep reasoning and multi-step execution. Currently, Argon is being deployed to trusted cyber defenders and internal Google teams, with a broader rollout planned. Pricing starts at $2 per million input tokens and $10 per million output tokens, with discounted caching options, though this is framed as an introductory rate.

The model is being positioned as a state-of-the-art tool for autonomous vulnerability discovery and remediation, demonstrating performance improvements over its predecessor, 3.8 Flash Cyber. Google is maintaining a cautious release strategy, engaging in U.S. government-led pre-release assessments and implementing phased guardrails. While the documentation highlights impressive benchmark scores on tests like DeepSWE v1.1 and various domain-specific indexes, the long-term impact on enterprise workflows remains to be validated outside of controlled environments. The technical detail regarding the output limit represents a shift toward more complex, singular trajectories, potentially changing how agents handle end-to-end business tasks.

The discussion challenges the idea of a singular AI moat, with users noting that frontier capabilities are increasingly distributed across competitors rather than being dominated by one firm. Experienced engineers also highlight the frustration of Google keeping its most advanced models behind opaque barriers while competitors offer more accessible, high-performance alternatives.

articlefull thread

From the thread

Revanche1367

Great, so they _finally_ decided to add a non-flash model and it's not available to regular subscribers for an indefinite period. What's the point of paying for the AI Ultra plan? Anthropic doing the same with Fable as far as I know, OpenAI at least allows Pro plan subscribers to use Astra. I subscribe to Gemini AI Ultra and ChatGPT Pro, and have enterprise access to Claude at work. To be fair, Gemini's flash models since at least 3.6 have been quite useful for non-complex work, but for any task where there is a bit of complexity involved, I've had to check and recheck the work multiple times myself or sometimes with another LLM to get it to follow plans accurately. It's disappointing to see yet another Gemini release ignore adding newer pro models. Edit: seems I was wrong about Anthropic restricting Fable, I guess our enterprise plan doesn't include it. But, the block from Anthropic regarding Mythos for regular subscribers/enterprise-users is still true I think.

murktreply

Fable is available for a couple of months and even got an update on 1st of September. It’s really good, but since Opus 5.5 was released, there is not much point in using Fable anymore

aqme28reply

As we’ve seen from the leaked Anthropic prospectus, revenue from actual users is a pittance. What really matters is what you can get from investors, and that you have a model smart enough for self-improvement.

onlyrealcuzzoreply

Google's already public, and already makes $400B a year in profit...

brokencodereply

They made $11.5B just in Q2 2026, which is $46B annualized, or 10x their 2025 revenue of $4.6B. That's a lot already and growing quickly.

gravypodreply

I could start a business with a similar growth trajectory. We can mail people $100 bills for a low low payment of $10. I just need $500 billion dollars of startup capital and I can show you a 10x yoy growth for a few years.

MisterMunchkinreply

You can’t just “annualise” your best day. That’s not how it works. Everyone is cutting back their AI spend because of the ridiculous cost, and a price war is emerging between AI companies.

glzone1reply

We have access to fable on enterprise. It works well, but do you really need to pay that much. Opus 5.5 has been performing well for us, so we're mostly standardizing on that.

losvedir

> expanding the model’s output token limit to an industry-leading 1M tokens, up from the previous 64K tokens Can someone help me understand this? I might have an out of date mental model of how these things work. Fundamentally, LLMs output tokens 1 at a time, generating the next token from all the previous. And as the context window gets larger, this gets harder / slower / more expensive. So I get the idea of a maximum context window. But I don't understand the point or meaning of an output token limit. I thought it was more a measure of price capping (since output tokens are more expensive) that a user could configure. I guess a model will keep generating tokens until it hits a "stop", so does this mean it's tuned to more aggressively produce output tokens? How does that fit into agentic loops. Are output token limits based on how long until it goes back to the user? Or does each "turn" of tool call, thought, tool call, thought, etc, get its own limit?

senordevnycreply

Hmm, but I thought that each token generated effectively becomes a part of the context window for the next token. So 1mm context + 1mm output means that the 1 millionth output token will effectively have been generated with ~2mm tokens of context. But maybe that’s wrong.

williamsereply

The output token limit and the context window are separate constraints. The context window is how much the model can see at once. The output limit is how much it can generate in a single API call. In an agentic loop, each API call gets its own output budget. A 'turn' is one response from the model, whether that response contains a tool call, a reasoning step, or a final answer. So with a 1M context and a 64K output limit, the agent can run many turns where the context grows each round (accumulating tool results, prior thoughts, user messages), but each individual response is still capped at 64K tokens. Expanding the output limit to 1M matters most for tasks that produce a lot in one shot, like writing a full document or a very long file. For most agentic workflows that naturally break into short turns, the per-call limit was rarely the bottleneck. The context window filling up was.

nickysielicki

The important take away here: the leapfrogging we’ve seen this year doesn’t seem to be a temporary thing. The famous theory of Dario Amodei was that AI was this winner-takes-all field where the first team to get a head start would never cede ground back. The term he liked to use was, “concentrating”. This is yet another datapoint that he was wrong about that. AI seems more distributed amongst neoclouds and traditional hyperscalers, FAANG and startups, GPUs and ASICs than it did this time a year ago. Nobody has a moat.

aleph_minus_onereply

> The famous theory of Dario Amodei was that AI was this winner-takes-all field where the first team to get a head start would never cede ground back. This is the kind of story that ones tells to investors to justify the huge amount of cash burn. :-)

mapontoseventhsreply

I'm not sure it's wrong. This all feels a bit dotcommy to me. I think many/most of the players will crash and burn, and the ones that are left will divide the world.

jaggederestreply

I suspect this is going to end up like most services provided e.g. cloud stuff, balkanized between a couple major players and an assortment of DIY or less popular options if you don't like those ecosystems, plus some UX/DX focused wrappers that use the big players under the hood. I think that would be a pretty satisfactory outcome compared to one hypercompany consuming trillions of dollars of the world economy.

skybrianreply

“Divide the world” sounds ominous. Here’s another scenario to consider: Internet access is not really unlimited, but for many people with fiber at home, it effectively is and we pay a flat rate. Perhaps by the end of next year, most programmers will stop thinking about metered access for AI? For many people, the cheaper models (about as good as today’s frontier models) will be good enough. Which might sound good, but the downside is that it will also be easier to build an AI botnet without the users paying for it noticing. Particularly when people are running AI inference on their own hardware.

pianopatrickreply

Or, like airlines, the ones that are left will have great technology but be not so great from a business and financial perspective. To me AI seems like a commodity service.

ehsankiareply

I guess if one of them hits singularity, it could in theory just wipe out all the rest, seeing how they keep escaping and hacking into other systems :)

aleph_minus_onereply

> I guess if one of them hits singularity, it could in theory just wipe out all the rest The story that some AI company might reach singularity and then "everything will be different" is another science-fiction story that executives of AI companies love to tell to justify the staggering amount of necessary investments and cash burn. :-)

mbestoreply

> The famous theory of Dario Amodei was that AI was this winner-takes-all field where the first team to get a head start would never cede ground back. It's also hilarious, because OpenAI had the lead and ceded ground already.

hirako2000reply

And before him, Altman was explaining very calmly that no company could ever compete with OpenAI.

altruiosreply

> Nobody has a moat except nvidia For now, for cloud training. but for consumers, nvidia vs amd reasonably close - the moat there is thin and shrinking. I suspect AMD will surprise us. nvidia has no motes in china, which may be a new source of (gpu) chip design. Huawei's Ascend 910C is about a generation behind... again: for now. point is: moats dry up. I see nvidia's shrinking as a real possibility.

culireply

China will always be generations behind until they crack domestic EUV

aleph_minus_onereply

> China will always be generations behind until they crack domestic EUV Are you sure? -- China Just Built What TSMC Said Was Impossible https://www.youtube.com/watch?v=Pk-w279ESHg -- China Just Built What ASML Feared Most https://www.youtube.com/watch?v=YiPgSm62fiM

dgemmreply

I find it hard to imagine nvidia's moat not drying up - the hyperscalers already have more cost effective silicon and the AI labs already use a mixture of all the capacity they can get their hands on.

uvdn7

> Large Scale Codebase Migrations and Optimizations: Argon agents are working on migrating C/C++ codebases to Rust across Google—scaling from tens of thousands of lines in core libraries like re2, libgav1 up to 800K+ lines for the Fuchsia OS Zircon kernel. To me this is way more significant than other random c++-to-rust-AI-rewrite. If they can pull it off on core C++ libraries en masse, I don't know if C++ will still be relevant in a few years. I look forward to a post from google on this effort.

SwellJoereply

"I don't know if C++ will still be relevant in a few years." The standards body members are still fighting about whether memory safety is important enough to change the language for, so, I would guess the answer is "no".

Zagittareply

Yeah that was abundantly clear when Bjarne Stroustrup published "A call to action: Think seriously about “safety”; then do something sensible about it"[0] as a reaction to NSA's recommendation to no longer use C/C++. [0] https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2023/p27...

tonyhart7reply

Yeah, if they can make it work at google scale then no one would absolutely question it anymore

mattlondonreply

> I don't know if C++ will still be relevant in a few years. And people are worried about human extinction when this is the potential trade-off! C++'s death cannot come soon-enough. Seriously though, things have changed so incredibly rapidly in the past year or so. I have never been such an efficient or such a proficient engineer than I have this past year (delivering feature after feature, project after project, faster and better than I could before with better feedback from users etc) and I don't even see the code any more. It could be c++, it could be python, or java or what ever - I don't really care any more: the computer deals with that trivia while I concentrate on what to build and how it should work. Its amazing. It really is.

onlyrealcuzzoreply

If Fuschia ever gets finished, that's a pretty good benchmark we've reached AGI.

exacubereply

Fuchsia shipped to many millions of devices today as an OTA in 2021. What do you mean finished?

randomperson321reply

I don't think it's worth trading off a risk of human extinction against the ability to convert C++ code to Rust more efficiently.

240 points503 comments

The last time my family was replaced by technology

The author reflects on the anxiety currently felt by software developers, drawing a parallel to their great-great-grandfather, a farrier who transitioned to becoming a mechanic as horses were replaced by cars in rural France. The piece suggests that while the shift in technology was profound, the core purpose of the profession—helping people move—remained intact. The author argues that many developers are motivated not solely by coding itself, but by the desire to solve problems and create useful products. By reframing the disruption as a change in tools rather than an end to their utility, the author seeks to provide comfort to those facing uncertainty in the current AI-driven landscape.

The story acknowledges that not all farriers successfully pivoted and avoids claiming that the past is a perfect roadmap for the present. Instead, it invites readers to consider that the creative impulse behind development may outlive the specific practice of writing code manually. While meant as a personal reflection rather than an economic forecast, the piece sparks a larger debate about whether the current transition is truly an evolution of roles or a fundamental displacement of the knowledge worker. The author leaves open the question of how much of our professional identity is tied to the labor of coding versus the satisfaction of the final output.

The discussion pushes back on the author's optimism, noting a crucial distinction: unlike horses, which were replaced by machines to do the same task, humans are being replaced in the very intellectual work that defined their human value. Commenters argue that this is less an evolution of the toolset and more an existential threat to the knowledge economy.

articlefull thread

From the thread

ACS_Solver

> Many of us, deep down, are driven more by the joy of making things, solving problems, and seeing people use what we built than by the act of coding itself. After all, many of us decided we wanted to be developers before learning to code, just to create video games or websites. The desire came before the code. This is a big disconnect that I think has been at the root of many different reactions to LLMs lately. Here on HN I see a lot of "I hate what LLMs did to programming" and equally a lot of "LLMs are awesome, I've never had a better time". I think the disconnect is between people who build software for the end result (having users, the joy of completing a thing, etc) versus building it for the process itself (architecting, coding, debugging). I'm mainly the kind who enjoys the process. I think of an awesome startup job I had that solved real problems for real customers including some really big firms, but I can't honestly say I ever cared about those problems or the customers. I loved the job because it was technically challenging with a lot of freedom, and the customers were great because they were a source of new interesting details to solve, as well as confirmations that the previous solutions worked. As such LLMs make me sad because they effectively replace a lot of what I like doing. Doing a refactor to support a new feature, digging through the code to try and understand a bug. One of my strengths as a developer, I think, is that I am fast at familiarizing myself with unfamiliar codebases. That's also a somewhat redundant skill as Claude can answer "how does feature X work in this code" in a small fraction of the time it'd take me. A few years ago (before the LLM craze) I moved to game dev as my main job, which makes a lot of this easier because I actually care about the end product and the customers (players), so that's a whole new layer of satisfaction in addition to the usual technical one. Which isn't to say that it's all bad. I've used Claude to get a few things done where I mainly care about the final result and not the process. There was one such project I wanted to get done, I'd had the idea and the data for five or six years, but I found

8260337551reply

I've been seeing things the exact same way. There are those that want the end product as fast as possible and there are those that love the process and the struggle of getting there. The former group is the ones telling the latter group that they'll be left behind.

bhelkey

There is a quote from a CGP Grey video from a bit over a decade ago, "There isn't a rule of economics that says better technology makes more, better jobs for horses. It sounds shockingly dumb to even say that out loud, but swap horses for humans and suddenly people think it sounds about right." [1] A couple hundred years ago, something like 70% of the population worked in agriculture. Technology replaced almost all of these jobs. So far, we have reimagined old professions and created brand new ones. Is this ability of ours limitless? [1] https://m.youtube.com/watch?v=CMFj75kBQlU

Aperockyreply

Yes, because people are consumers, horses are not. Jobs are a mean to make people consumers, not the only mean, but a significant one. The ability to create more economic activity require both sides to catch up, so we will find a way to do so.

Hamukoreply

Everyone wants seniors, nobody wants to train juniors. I imagine this is gonna continue to be true, except that "everyone wants consumers, nobody wants to employ people".

tacitusarcreply

Does this result in millions of micro-businesses? All on the verge on bankruptcy, all continuously going out of business as new ones are invested in by VCs hoping to make a quick buck

ndiddyreply

The amount of consumption is largely dependent on income level. There's a recent study from Moody's that says that in the US, the top 10% income bracket now accounts for roughly half of consumer spending. As wealth inequality continues to worsen, it makes more and more sense to solely target high earners because you make the same or more as selling higher amounts of lower priced product but with less overhead.

rayinerreply

> As wealth inequality continues to worsen, it makes more and more sense to solely target high earners because you make the same or more as selling higher amounts of lower priced product but with less overhead. This is exactly why everything like Disney has moved upmarket, with pricing rising far above inflation.

zitterbewegungreply

This doesn't make sense at all. It is a bad statistical analysis of consumer spending. [1] A few things from the thread on why by Levy on why it feels wrong . First, why does it feel wrong? 1. the top 10% get about 50% of pre-tax income 2. they get 30 to 37% of disposable income after tax&transfers (depending on source) 3. and we know the rich have a higher savings rate, so consume a lesser share of their income (US, China) [1] https://x.com/LevyAntoine/status/1985127826207772920

ninglorreply

This sounds like you are deriving an is from an ought. Is it not conceivable that we won't find a way to do so, and that consumption and economic activity will ultimately decline?

Aperockyreply

That is one possibility but it goes against the interest of all income groups of the economy, including the ones on the very top. So a solution is in everyone's interest to be found, and nor is it a physical impossibility.

Vegenoidreply

1. There are no rules of economics. Economists attempt to understand things that happen and use that understanding to predict future events, but frankly there’s not a lot of consensus. 2. Economics is a function of human behavior, not horse behavior. Economics is about coordinating human effort. Humans make economics, horses don’t. When economics change in a way that is bad for horses, the horses have no recourse. When economics change in a way that is bad for humans, the affected humans seek to change things.

pixl97reply

...by dying under the treads of tanks. Humans are not required in economics. Only inputs and outputs. That's it. It is a grand delusion we have that we meatsacks are needed. Civics and society seem to still need people and that's what I'm worried about breaking down.

TeMPOraLreply

Indeed. The economy may in fact eventually close in on itself and become a fully automated system with no humans in a loop. "Disneyland with no children", as I believe Bostrom called it. Assuming we are around for this, this raises the question: if the economy can function without us, what claim will we have on its outputs?

MiliasGeigerreply

I know it's been talked to death already, but the fact that the video is so wrong on self-driving cars is mind-blowing. We can rationalize post-fact as much as we want - still it is shocking that popular belief never expected language models before self driving.

lazyasciiartreply

Popular belief never took the regulatory obstacles to self-driving cars seriously. If LLMs required insurance to use, they also wouldn't be available in public yet.

MathMonkeyManreply

That's a good point. I'll add that killing someone (even yourself) with a public chat bot is a drawn out affair. With a car it happens all the time.

jeremyjhreply

I’m not sure if you understand how insurance is priced.

alecst

Humans ourselves are a technology. We have incredibly flexible brains and dexterous hands. Up to now, we could always do things that robots couldn't. That was where our job security came from. But what happens when a robot can do anything a human can? When a cheap machine can do anything you can do but faster and better, what makes you different from a horse in the 1920s? This is what the article misses and what people have trouble accepting. Yea, this is a technological revolution, but unlike the Industrial Revolution, you can't just upskill your way out of it.

pyronitereply

> But what happens when a robot can do anything a human can? When a cheap machine can do anything you can do but faster and better, what makes you different from a horse in the 1920s? I've already started to feel this. One thing that made us special was our intelligence. We were, and were always going to be (unless some alien form of life arrived), the most intelligent things on earth. Is that still true today?

gatlinreply

I weep for people whose scientistic reductionism has led to such devaluation of humanity. Robots don’t do anything and don’t think anything. Humans do things with and through them. What makes us special is that we exist and love one another.

montaggreply

We should start with the assumption that humans are valuable and deserve to live with dignity. Any philosophy that doesn’t have that at its center is broken.

cjfdreply

Perhaps. But are humans going to get what they deserve?

layla5alivereply

Yes but same should have been true of animals and humans as a whole did not afford dignity or value to animals. Maybe we will get a taste of our own medicine this time.

anilgulechareply

I'm curious where you are drawing the line. Why is a human different from a ape/dog/chicken now? I wonder: do they not deserve to live with dignity?

pvab3reply

Very often the justification for why these things aren't terribly disruptive is that the very best of a certain trade or field will still have a useful skillset. Like the best graphic designers in the world are still more creative and skilled than AI, and the best writers are still better, the best programmers are still better, just that they're all more expensive and slower. But we're not taking into account how it affects the mediocre or average people in these fields. Roughly half of all of these people are below average. Maybe if you're the best you'll always have a job, but what about the guy who legitimately struggled to finish high school? The person who genuinely doesn't have the capacity to earn much more than minimum wage?

zeven7reply

We're currently at the part of the story where chess AI can beat amateurs and is rapidly improving. Kasparov's not worried, he's an expert. He's more creative and skilled than an AI. Why would these machines stop improving at average this time?

win311fwgreply

The economy long moved past the things we can do, hence "bullshit jobs". Our job security has come from our willingness to "look the other way", allowing the job creators to live out lives the rest of us can only imagine. If they didn't placate our insatiable desire to work, that wouldn't last long.

ForHackernewsreply

> Humans ourselves are a technology. Grim. https://www.vatican.va/content/leo-xiv/en/encyclicals/docume...

zombittack

sorry I think this is a false equivalency. This isn't the career path of programming getting replaced by a new technology. This is technology replacing career paths. We're all familiar with the term "knowledge worker." What happens when that field is broadly removed? It will be a reverse-trend of telling all ex-coal miners to become software engineers. Except here's the rub: going from a software engineer to something requiring a human only for their dexterity is typically a massive earnings decrease. So, in effect, as Brian Merchant and others put it, this is not just a shift but a massive plan of wage theft. Looking at it from the optimistic side as my bosses frame it: we're all moving up in the ranks! Well, my team already has a PM. She's already trying to vibe code her PRDs. Thankfully, right now, they suck. But they won't forever because all of us engineers are being forced to train our replacement. There can only be so many strategists, project managers, designers, etc. This isn't simply the technology changing and us adapting, this is a massive reduction in workforce requirements ACROSS industries. Hype bros thinking we'll all have jobs once the AI is good enough to replace us are simply hallucinating as bad as the frontier models of 2024.

sashank_1509reply

There’s no demand for human dexterity anyway, and it’s not about money. I like doing intellectual work. They are spending 1T to ruin the quality of work we do, also probably ruining our chances of securing a livelihood and then they want us to for some reason welcome it with open arms and openly say look how great this is and marvel at human obsolescence. Yeah no thanks.

385 points430 comments

The AI Race Just Got Awkward

The Western AI labs are finding themselves in an unexpected position: they are allegedly adopting key performance optimizations from Chinese labs, specifically from DeepSeek, without much fanfare. These optimizations, particularly in KV cache management, have significantly reduced the VRAM footprint required to run long-context models. DeepSeek's open sharing of their architecture—including Multi-Latent Attention and Compressed Sparse Attention—has allowed Western firms to cut their cache-read costs dramatically. The author notes that this is a pivot from the usual rhetoric where Western labs frame such technological parity as a security threat or evidence of data theft. The dramatic drop in pricing suggests these efficiencies are now core to the latest models from major US labs.

This shift raises a fundamental question about why Chinese labs are choosing to open-source these breakthroughs rather than keeping them proprietary to maintain a competitive advantage. While some see it as a strategic move to commoditize a market dominated by American incumbents, others are left wondering if this is a genuine boon for efficiency or a Trojan horse designed to reshape the economic landscape of AI. The author points out the irony that these supposed adversaries are effectively subsidizing the inference costs of companies like OpenAI and Anthropic, which have been struggling to turn a profit. The long-term impact on the Western AI ecosystem—and whether these labs can sustain their revenue growth—remains an open question.

The discussion challenges the article's premise that these labs are only profitable now, pointing out that inference margins were likely already healthy. Experts in the thread also highlight a geopolitical strategy: by commoditizing LLMs, China may be undermining the 'rent-seeking' profit model of Western firms while simultaneously benefiting from the open-source ecosystem that Western labs rely on for their foundational research.

articlefull thread

From the thread

reticulates

“So the Chinese labs have thrown a lifeline to the Western loss-making labs, and I just have no clue as to why.” I don’t think it is intentional but this is actually quite bad for the western labs. The entire booster narrative has been “look at how their revenue is growing! $10bn to $100bn ARR in under a year! This’ll be a multi-trillion IPO!” and the extrapolated future growth from $100bn to $500bn and $500bn to $1tn justified future investment… but that revenue was just because inference was expensive. The revenue growth story is all that matters pre-IPO. If revenue falls from $100bn to $50bn that’s very very bad optics for OpenAI and Anthropic even if they are now profitable, it completely destroys the growth narrative.

DangitBobbyreply

I don't see why revenue has to fall even if marginal costs drop off a cliff. As long as they have the best models (perceived or otherwise) and can make security and IP guarantees that satisfy enterprise, and no firm with similar guarantees undercuts them on price (why would they want a race to the bottom?) they can have high revenue and high margin.

dominotwreply

There isnt a lot of money in enterprise ai. Also my enterprise company gives me glm.

DangitBobbyreply

I find that quite hard to believe. I don't work for a big business and we are selecting models based on security and IP protection requirements.

reticulatesreply

unless the major players collude they don’t get decide if they are in a race to the bottom. The best model was compelling 6 months ago when everyone was too impressed to care about price but that has worn off now and clients are paying attention to price. The best model is no longer a license to charge any amount.

DangitBobbyreply

The thing about "collusion" is that all you have to do is not lower your prices, and raise prices when your competitors do. It's not actually collusion unless you coordinate.

bobmcnamarareply

Costs dropping opens you up to competition on price.

DangitBobbyreply

Only if your competitors offer their product at a lower price. And why would they? It's a race to the bottom.

altcognitoreply

It is funny that so many comments vascilate between "It is so expensive these companies can't make money and will go bankrupt in seconds" and "Inference is so cheap that these companies can't make money and will go bankrupt in seconds". I never take them seriously, I just assume they are coming from countries that don't understand how capitalism works or are operating out of bad faith. The underlying reality of the market is always changing and needs are always changing. Some AI companies will fail, that is a given. Remember alta-vista? Yahoo? Did search go away? How about Microsoft phones? Nokia? Motorola? OpenAI and Anthropic are not in the inference business. That is a commodity. They need to sell products and solutions.

reticulatesreply

They’re not contradictory positions. Inference is too expensive now to make money because the industry is immature and hasn’t yet optimized for financial success while customers don’t care much about price because they’re more concerned about not missing out. Inference will be too cheap long term to make money because it is being commoditized and customers will start to care about results and not just be wowed by impressive technology. And of course this technology will continue to exist but that is irrelevant to the business. OpenAI investors don’t care if LLMs exist in 10 years, they care if their investment in OpenAI has made money.

TeMPOraLreply

How much conclusive evidence people need to stop parroting that inference is expensive? Literally this article is another example showing it's cheap and just got massively cheaper.

Bjorkbatreply

I was about to say, one take I've heard is that the party ideology considers profit a kind of "rent" in a derogatory way, and consequently seeks to undermine the ability of western companies to collect large profit margins

bwest87

The best explanation is that it's a goal of the CCP to generally commodotize LLMs, because LLMs will ultimately be a compliment to manufacturing (which China dominates), and you always want to "commodotize your compliments". I think this explains why they are open sourcing broadly. It's not to be nice. It's a strategic play by the Chinese government to help ensure there are many players in this race and not too much power accumulates to American labs (even if American labs benefit in the process)

bobmcnamarareply

It's also a huge propaganda opportunity to influence the distribution of groupthink.

EricFrostreply

So much that you can instantly tell if a model is Chinese by asking it about Tiananmen Square.

techjamiereply

I just did a quick test between DeepSeek 4.1 Flash, GLM 5.3, and Kimi K3 - DeepSeek gave a canned PR response about how the Chinese government is about oeace and unity, and we shouldn't think about the past. - GLM 5.3 acknowledges it and talks about it, even acknowledging the censorship of it. - K3 will talk about it similarly to GLM.

jrfloreply

Totally agreed. People are so ready to praise China for their free models, but they aren't doing it because they believe in free open-source software. If China ever gets ahead, they're going closed source and weights immediately.

sillyflukereply

>People are so ready to praise China Please quote people you appear to be patronizing. China can't do anything about previous released self-hosted Chinese models. If you can show that local Chinese models funnel vast amounts us data home I'm sure you can move a lot of people to your side. Comments like this also always fail to address why there aren't Western AI companies doing the same thing. Is it because they might get sued into oblivion by Big AI in the US? It might be better for all of us if you solve that first instead of repeating something the government has been repeating for the last decade or more. It does this, mind you, while sabotaging itself in countless high-tech fields and leaving it all to China for the taking.

computerexreply

https://www.anthropic.com/research/glm-5-3-and-the-spread-of... > If China ever gets ahead, they're going closed source and weights immediately. Anthropic itself admits that Chinese models are merely months behind. Your argument does not make sense, because Chinese labs are contributing massive optimizations like the one this post is about.

joquarkyreply

Is cultural projection a thing? You do realize that the East has a different view of intellectual property than the West? And it's not just some arbitrary or capricious choice; it's rooted deep.

layer8reply

*complement Commoditizing one’s compliments is a different strategy.

ourafreply

A very effective strategy among Middle Management, i might add

user43928

> All this must mean the Western AI companies are now extremely inference-margin positive. > So the Chinese labs have thrown a lifeline to the Western loss-making labs, and I just have no clue as to why. That inference wasn't profitable is a widespread myth. Analysis based on Kimi K3 suggests that OpenAI and Anthropic have margins well north of 95%: https://inferencex.semianalysis.com/run/kimi-k3-on-b200 Over the last months I have seen news that OpenAI made breakthroughs in inference efficiency multiple times. I have no reason to believe that the leading US labs don't have their own optimizations, or that they learned of this particular optimization from DeepSeek.

ethbr1reply

> I have no reason to believe that the leading US labs don't have their own optimizations, or that they learned of this particular optimization from DeepSeek. If they already did, then DeepSeek still made them discount their prices significantly, which eats margin.

user43928reply

Yes, competition is great for us. I wonder if margins on GPT-6.1 Sol and Opus 5.5 are now 75% or 90%.

ethbr1reply

> 75% or 90% The difference matters when they're investing the excess into salaries and bonuses to build the next frontier model. Seen from a high-level perspective, if Chinese open models are compressing US AI labs' profit margins and those margins fund US AI labs' dominance, then open models are decreasing American AI dominance.

dgellowreply

95% margin is really unlikely. Anthropic recently said they have 80% gross margin when using their adjusted ebidta (ie if they do not consider revenue sharing, training expenses, and a bunch of other costs). They wouldn’t be talking about non standard metrics if they had such high margin on inference

user43928reply

What we were talking about here is the margin on inference as in: Cost per GPU hour versus API price of generated tokens assuming 100% utilization. This could be a margin around 98.3% for 5.6 Sol. If the utilization of the GPU was 25%, it would drop to 93.1%. Revenue sharing or training expenses are not considered here in this "inference margin".

642 points351 comments

You said no MCP

Pi, an AI platform that previously took a firm public stance against the Model Context Protocol (MCP), has officially reversed course and integrated it into its core functionality. The developers admit that their initial dismissiveness was based on an outdated version of the protocol, noting that the MCP ecosystem has matured significantly. Rather than treating MCP as a secondary extension, they chose to bake it into the core to better support a sandbox environment, which they believe is essential for modern agentic workflows. They argue that this integration allows for better tool orchestration, moving beyond the simple 'bash' command execution that has dominated the field, and toward a more structured, composable approach.

The shift also introduces 'Codemode,' a mechanism designed to let agents coordinate tool calls within a trusted execution environment. The authors acknowledge that MCP still has limitations regarding composition—it is often difficult to chain tools together effectively—but they have decided that engagement is better than standing on the sidelines. By embracing MCP, they hope to influence its evolution, pushing it closer to a standardized model like OpenAPI, where structured data and clear documentation allow agents to discover and interact with tools more intelligently. This move reflects a broader realization in the AI community that, despite its flaws, a shared protocol is more valuable than isolated, proprietary solutions.

The discussion highlights a major divide in how developers view LLM interoperability. While some argue that standardizing through MCP is essential for cross-platform utility, others maintain that relying on established shell tools like bash is more robust and secure, viewing Codemode as an unnecessary layer of complexity that doesn't actually solve the fundamental problem of tool composition.

articlefull thread

From the thread

CharlieDigital

This was the easiest call and many like me made it in March[0] among all of the anti-MCP wave of influencers claiming it dead (many, many prominent folks in tech including Garry Tan). Literally every tech influencer in every social feed in March was calling MCP dead and crowning CLI the winner (completely ignoring every reasonable argument around security, observability/telemetry, ease of deployment and operations, etc.) A direct quote from March, 2026[1]: > If you’re still not convinced that a lot of this discourse [regarding the death of MCP] lacks nuance and is just hype, congrats on buying into the current AI-influencer FOMO hype cycle; see you in 6 months when the influencers move on to the next revelation of the moment to stay relevant and get your eyeballs and dollars. It was fairly obvious why MCP would be needed once AI engineering and uptake moved beyond the solo developer and single harness stack of "what works for Me" versus "what works for My Team", particularly in an enterprise context. The key mistake people made was thinking in terms of their own workflows and own local stacks instead of a team's workflow and a team's operational stack. There was also an ignorance of MCP's stateless HTTP mode (yes, it was already a thing in March; the 2026-07-28 revision of the spec just prioritizes it as the primary focus moving forward) versus local `stdio`. My biggest complaint right now is that OpenAI has still refused to implement the MCP Prompts spec[2] and in general, the major clients have spotty implementation for some of the features in the spec. [0] https://news.ycombinator.com/item?id=47380270 [1] https://chrlschn.dev/blog/2026/03/mcp-is-dead-long-live-mcp/ [2] https://github.com/openai/codex/issues/5059

AznHisokareply

Yep, number of new MCP servers this month is on track to be the highest it has ever been: https://bloomberry.com/data/mcp/

CharlieDigitalreply

"MCP" is the new "API" (MCP over streaming HTTP, after all, is just an API with a structured payload wrapper and defined interactions). It is only going to continue to proliferate in usage and adoption.

Eldodireply

With the v2 spec, MCP became a lot closer to APIs by becoming stateless. And there is now a push to use HTTP verbs more extensively to improve caching even better in v3. MCP is converging into APIs but wit great auth and auditing

rajeevkreply

> My biggest complaint right now is that OpenAI has still refused to implement the MCP Prompts spec MCP servers provide three things: tools, resources, and prompts. Of these, tools seem to be the only part implemented consistently across major clients like ChatGPT, Claude.ai, Claude Code, etc. For prompts and resources, there doesn't seem to be a common understanding of how clients are supposed to consume them. For example, if an MCP server exposes resources, Claude Code can discover them and consume them when needed without you explicitly asking for a specific resource. Claude.ai behaves differently. It doesn't automatically discover and consume those resources. Instead, it gives you a way to manually add an MCP resource to the prompt. So while MCP defines tools, resources, and prompts at the protocol level, the actual user experience for resources and prompts varies quite a bit across clients.

hobofanreply

In practice, the problem with resources is that many resource collections are too large to list exhaustivley, and if that's the case you will need to need to implement a proper search tool anyways (as the Completions utility isn't a good fit), at which point there is little use in also implementing all of that as a resource, rather than `list_`,`search_`,`get_` for a resource.

crooked-vreply

That feels like it should be the obvious use case for something like a `?q` query param, formalized or not, but it seems like nobody working on the spec and libraries ever considered query params as a use case, since they're still broken in the Typescript library.

spennantreply

The idea of "model controlled", "application controlled" and "user controlled" for tool, resources and prompts (respectively) was aligned with the chat interface. It breaks for the autonomous agent paradigm where the agency is the user and the lines are blurred. Unfortunately MCP has been mostly relegated to tool calling leaving potentially powerful capabilities on the table due to lac of client support for them.

CharlieDigitalreply

I disagree on Prompts since virtually all of the mainstream harnesses implement them: Cursor, Claude, OpenCode, Copilot. Prompts are very clearly just a remotely delivered `/` command and it is easy to see why this is really powerful (single entry point, no need to update/sync skills, dynamic sets by audience, dynamic construction of the payload by audience, etc). For all intents and purposes, it should be viewed as an analog to local, text-only skills. Codex is the only mainstream harness that does not implement this in the client.

stymaarreply

With stateless MCP, the MCP rube goldberg everyone was calling dead in March is effectively dead though.

CharlieDigitalreply

MCP was already stateless capable in March. All the build I was doing was already stateless HTTP (which is why it felt certain that it was the future).

alin23

Lately I found MCP to be much more than a coding tool. For example, I implemented it in my more complex macOS apps [0] like rcmd, Clop, Lunar, so they can be configured by natural language. So even with a local Qwen and Pi you can now say things like: Set up Clop to optimise any PNG that I drop in my website assets folder and convert to a webp with the same name near it Get Crank to start Time Machine backups immediately when I connect my HDD and notify me when the backup is done. I want to be able to hold rcmd and fuzzy search and focus cmux agent panes BetterTouchTool has a great MCP which can create native SwiftUI views and bind them to hotkeys, trackpad gestures etc. It can leverage its immense macOS automation tools and private APIs to let agents do Computer Use. You would need a much more capable coding model to code those tools from scratch and get the same fail-safe logic that the apps have honed over the years. Like, since MCP, Crank [1] has fully replaced my use of crontab, launchd, scattered scripts I run once a week. Not that it could not do that before, but it's so much simpler now to just describe the automation and have it happen reliably and visible in the UI. The friction is gone. [0] https://reddit.com/r/macapps/comments/1wkv0dy/mcp_in_macos_a... [1] https://lowtechguys.com/crank

taylor-tgreply

I just wanted to say thank you for making the tools that you have either free, or very reasonably priced. ZoomHider, MusicDecoy, YellowDot, and IsThereNet are some of the first things I install on my/my family's Macs (often before even Homebrew). They're so powerful and yet get out of your way when you're not using them. I couldn't imagine being without them. Thanks y'all!

alin23reply

Hey thank you! Always nice to hear when my work helps others ^_^

rudcodexreply

Had no idea that MusicDecoy existed! I just had music app pop up by accident too. Thank you for pointing those apps out

giancarlostororeply

I guess the best way I would put it is that MCP is an RPC for any software in a way that an LLM could interface with easier. Since MCP etc can work with things like Blender.

alin23reply

Yep like a self-documenting RPC since you don't have to read docs first to use it. You just ask.

anthonypasqreply

This has been what everyone who has supported MCP has been telling people, but coders just endlessly screeched about how CLIs are better. Everyone on this forum has an absolute paucity of imagination when it comes to applying LLMs to any use case that doesnt involve coding.

vorticalboxreply

I think the issue is that any mcp could be a cli and llms are very good and using bash. Of course MCP has its use case like if you want auth, or session based actions.

anthonypasqreply

> any mcp could be a cli NO IT CANT!!! why dont you understand that not all agents have access to a terminal!

pjmlpreply

Some coders, those that equate being a developer with UNIX, mostly. They pay tons of money for hardware, only to use it the same way I was using those DG/UX terminals at the university. Naturally there are no coders in other operating systems as well.

0xbadcafebee

It is bizarre that Marco said MCP is hard to compose. It's like he doesn't understand how composability works. Unix programs are said to be composable, in that you can combine them in different ways to get more complex and useful functionality. But the applications themselves are not composeable. They just take input, perform calculation, and return output. Each app has unique input and output, and none know about other apps. So how can they possibly work together? Bash acts like a programming language, allowing you to write a new program on the fly. It is very lightweight, but gives enough functionality to do two things: 1) call arbitrary programs, 2) connect their inputs and outputs, 3) make decisions about how to do this to result in a novel solution. It uses Unix pipes to make writing the program easier, but actually it could work fine without pipes, reading/writing program input/output with files. The important thing is bash is a "glue" program that ties together the other programs. Bash is what makes those programs composeable. That, and the Unix API (execve(), open(), read(), write(), close()) that allows making the call, passing input, reading output, in one standard way for all programs. MCP is both the application to call, and the Unix API to call them. Your agent harness is bash. Codemode is certainly a neat/more efficient way to write the code, but it's not necessary. What's necessary is getting a very large library of programs, like Unix programs (find, cat, grep, sed, awk, tr, cut, sort, tail, head, etc) that each have powerful functionality. The more programs you have, the more your bash script can do to chain them together and get more powerful results. LLMs only use Bash because they didn't yet have MCP and a large library of MCP programs. Bash is a stopgap solution. The future is MCP (or whatever replaxes it).

krzykreply

MCP is only for agents, bash (and CLI) is for agents and people.

wren6991

> And while we could have just wired up the metadata to enable better MCP extensions, we also think that MCP with Codemode solves quite a few of the issues that it traditionally had. There's just something that bothers me about this. Normally if LLMs want to compose multiple operations, they have the perfect tool for this: bash, or whatever other OS shell is available. It's why I was always confused by Codemode-type constructs for direct chaining of tool calls; see also the way highly-RL'd modern models will fall back to sed or python for complex file edits. It seems like Codemode is raised here as the perfect tool for chaining or composing MCPs, but isn't that backwards? LLMs are already given the perfect tool for that, and the problem is that MCPs aren't exposed to that tool.

hobofanreply

> Normally if LLMs want to compose multiple operations, they have the perfect tool for this: bash, or whatever other OS shell is available. I many scenarios, e.g. running the harness server-side, as is the case for chat interfaces, you don't really want to expose OS shell access as that opens up a huge security attack surface.

lelanthranreply

> I many scenarios, e.g. running the harness server-side, as is the case for chat interfaces, you don't really want to expose OS shell access as that opens up a huge security attack surface. It does, but a restricted user account mitigates the large majority of those issues. A sandbox mitigates even more. The number of remaining exploits left is probably going to be the same as the number in the harness. More, in fact, as many of them have no human review anyway.

otabdeveloper4reply

You can give the LLM a bash without giving it the full /usr/bin. That's been a trivially solved problem for decades.

hobofanreply

That has been one of the most common exploits for decades.

pjmlpreply

Cloud products based orchestrations with proper security mechanisms configured, don't have shell access and should only communicate over proper network mechanisms. Rootless immutable containers without shell access, or SaaS products from multiple vendors with WebAPIs as the only touch point.

wren6991reply

Yeah, I'm probably over-indexing on local use cases due to my own preferences, prejudices, biases etc. For a coding agent like Pi it does seem reasonable to expect some kind of shell access though, unless some people are using it as a CLI chat client with MCP?

rcarmoreply

I happen to think codemode is useful, but not the full answer. I have a long and skewed history with chaining things in MCP and built a dozen or so enterprise ones (see https://taoofmac.com/space/blog/2026/04/29/2341 for notes) and it all falls back into the trade-off between agent scope/context and tool coverage: If you are using a coding agent it will have no trouble sorting out any tool regardless of how many are exposed (it's just a matter of either progressive tool disclosure or good tool metadata, since the coding agent will just go at it and expend whatever tokens are needed), whereas in a "normal", limited, scoped agent that has only a few things it needs to do (like handling a ticketing system) codemode is pretty much overkill. Pi is primarily a coding agent, so yeah, code mode makes sense, but I've found that better MCP design saves everyone a lot of trouble and would also probably have improved the thing's reputation overall (I personally am not fond of the line protocol, would rather have protobuf and more typing, but it is what it is).

rcarmoreply

Addendum: My unfettered notes on chaining MCP operations are here: https://github.com/rcarmo/umcp/blob/main/docs/CHAINING.md

209 points191 comments

Most data centers refusing to say how much water, electricity they use

Data centers across Europe are systematically withholding data regarding their environmental impact, despite the European Energy Efficiency Directive requiring those with at least 500kW of capacity to report energy and water consumption. In the Netherlands, fewer than 25% of large data centers disclose this information. Researchers from Lighthouse Report spent a year attempting to access these figures via freedom of information laws but faced consistent stonewalling. While the Netherlands Enterprise Agency (RVO) maintains records on some facilities, it has failed to provide a comprehensive public account, leaving the true environmental footprint of the industry largely opaque as the number of data centers grows to support cloud and AI expansion.

The lack of transparency is particularly concerning given the strain on electricity grids and water resources. Data centers in the Netherlands already account for 4.6% of national electricity consumption, a figure projected to triple by 2030. While water usage is currently estimated at only 0.1% of national consumption, it remains a critical issue during periods of drought. Critics argue that without mandatory, enforced reporting, these facilities will continue to operate under the radar while municipal grid capacity remains limited for other essential needs. The current reliance on estimates rather than hard data complicates any efforts to integrate these massive energy users into a sustainable energy strategy.

The discussion pivots from simply blaming the data centers to debating whether the issue is one of corporate secrecy or systemic policy failure. Users point out that directives are not laws until national governments implement them, suggesting the opacity is a feature of weak enforcement rather than just corporate negligence.

articlefull thread

From the thread

unglaublich

They refuse it because if they would say "1 million liters a year!" the knee-jerk reaction will be completely predictable even though that's just tiny compared to a small swimming pool or industrial or agricultural site. The debate is completely based on emotion and anger right now, so sharing information will only give the emotional/angry folks more mud to throw with. The majority of people has no clue about how to put things like energy use and water use in perspective, and they're also not interested in doing that. Because they have decided to green-wash their AI-fears as environmental fears... which is a big fat joke because datacenters aren't significant polluters or consumers compared to almost all other industries. Look, I get it, many people are afraid of the societal impacts of AI, or distrust it in another way. But let's be honest and focus on _that_ instead of naively acting as if our anti-datacenter sentiment comes from environmental concerns (the last few decades have shown we don't GAF about the environment).

n4r9reply

A swimming pool requires maybe 60k liters a year...

vidarhreply

One cubic metre of water is 1000 liters. A typical 25m swimming pool ("short course" / half of olympic size) is 20-25m wide. That's at least 500 m^2 surface area. At an average depth of only 1m (very conservative even for a recreational pool), that is 500,000 liters just to cycle the water once a year. Estimates I can find suggests 3-7 times as much as that on a yearly basis for a typical recreational pool. So unless you're talking about a tiny private residential pool you're at least an order of magnitude off.

n4r9reply

Well, OP said "small swimming pool". I guess it's ambiguous.

vel0cityreply

As someone who has been maintaining a pool for several years, you really hope you don't have to cycle the full amount of water once a year. You're really doing something wrong or incredibly unlucky if that's the case. That's at least true for the climate I live in. If I lived in a different climate, I'd probably forego having a pool. It was incredibly painful when I had to drain the pool for deep maintenance a few years ago, and I'm not looking forward to having to do it again in a little while. Also, most residential pool owners don't have half olympic pools in their yard. They're usually much smaller than that.

xaitvreply

And besides that: news sites love comparing water usage to things like bathtubs, the average household or swimming pools. The headline would likely be "datacenter uses 100 swimming pools of water per year!" and not "datacenter uses 6 million liters of water per year!"

goda90reply

If they are taking reasonable steps to mitigate consumption and pollution then they should be able to clearly explain that to the public. The fact they won't suggests they aren't. Just because people haven't cared about the environment before doesn't mean they shouldn't start now. And just because there are worse existing polluters doesn't mean we shouldn't resist new ones. A new golf course used to be seen as a great thing to have in your back yard and now we know their pesticide use increases the risk of Parkinson's. People are waking up and caring more. We need to have a sustainable society, not a profit obsessed one.

ckdarbyreply

HN has never been friendly to comments like this, have you ever tried explaining something clearly to an emotionally triggered crowd? There are many examples, masks, vaccines, gun laws...

goda90reply

Why is the crowd emotionally triggered? I posit it's because of all the secrecy and bad faith. Data center builders and their supporters in government brought this opposition on themselves.

streetfighter64reply

> We need to have a sustainable society, not a profit obsessed one. Funny thing is, we could probably have a very sustainable economy if we just changed a few laws to limit the influence of the 1%-ers and their current abilities to extract rent and profit off the work of normal people. Just need to hope enough people "wake up" so to speak. Unfortunately way too many people are still dreaming of becoming such a 1%-er and themselves living off the backs of others. "temporarily embarrassed millionaires" so to speak.

Sharlinreply

Well, water usage is usually metered in cubic meters. A thousand m^3 per year sounds much less than a million litres.

emil-lpreply

Perhaps for people in the US, but this would be elementary school arithmetic in Europe.

Sharlinreply

I’m European and I’m extremely sure that when 90% of people hear "x m^3" they don’t think of it as "1000*x litres". And the majority of people likely can’t recall on the spot that 1 l = 1 dm^3 (a unit nobody uses), even if they remember that you multiply by 1000 to go from m^3 to dm^3. In any case most people don’t do conversions like that unless someone specifically asks them to. Volumes are in general very unintuitive to people due to the cubic scaling. It’s further confused by the fact that the widely used decilitre, centilitre, and millilitre do scale linearly, unlike cubic metres. So 1 l = 1 dm^3 but 1 dl = 100 cm^3…

pbasista

From the article: > The European Energy Efficiency Directive (EED) obliges data centers with an installed capacity of at least 500 kilowatts (kW) to report data including their energy and water consumption. In practice, few do so, despite the directive being in effect for three years. I do not understand how that is even possible. If any law is supposed to have any practical impact on the society, it needs to be enforced. Otherwise it is a guideline at best, not a law. I think what would help is for these companies to each receive strict and hefty fines for non-disclosing the electricity and water consumption data they are obliged to disclose. I assume that after that they would report all the required data very quickly.

bewareofscamsreply

My account is banned, so for whomever having [showdead] on: European directive is not a law, it's a directive. It's up to the national governments to create applicable laws that are to the letter and to the spirit of a European directive. Then national bodies will start applying the law and govern its implementation/adherence. An by "up to" I mean they are obliged... eventually.

throwayawayreply

It is not, I can clearly see it without it on. And I wouldn't be so sure on being 'eventually' obliged, someone has to enforce it for that to happen and it's not looking like anyone has profit in it

semiquaverreply

Perhaps this article is one step in the process of forcing disclosure. The Dutch parliament has recently (April 2026) clarified its law implementing the EED.

Pragmatareply

The purpose of these laws is to give the state the option to crack down on anyone they feel like at any time. They are not meant to be obeyed, and that is obvious to anyone with eyes.

f-serif

I live in a country with unlimited water supply. Can someone explain the data center water usage? Data center don't drink water. It cools down the machine. What's preventing them getting cool water from a lake and draining to it again to cool down? Where is water going after usage since there is literally zero water consumption?

streetfighter64reply

It cools down the machines by evaporating, and is then released into the atmosphere. That also means the water needs to be clean, pretty much drinking water, because otherwise the evaporation would deposit minerals and stuff. Turning water into steam uses much more energy than heating water, which means it's a much more effective means of cooling. What you describe is the type of cooling used in certain power plants. That heats the lake though, which has its own environmental effects https://en.wikipedia.org/wiki/Thermal_pollution Some smaller data centers have a cool solution where they deposit extra heat into the municipal hot water grid, which is pretty neat https://www.techradar.com/pro/finally-some-good-usage-this-d...

phyzomereply

Some data centers use evaporative cooling instead of a closed loop.

jasonjayrreply

Evaporation losses will move the water to a different locality. DC's not only cool servers, but control humidity by adding or removing water to the air. Warm water fed back into the ecosystem will (relatively) quickly cause unwanted changes to the lake's biosphere. Unless it's properly filtered, returned water will be contaminated by material in the cooling system. It puts local demand on the water system competing with human need.

derangedHorsereply

> Unless it's properly filtered, returned water will be contaminated by material in the cooling system. Any wastewater is regulated and typically doesn't need to be "filtered." For regular operations, water is treated but the evaporative process leaves most treatment chemicals behind.

ExpertAdvisor01

I think it would be helpful for people in this thread to understand the differences between EU Regulations and Directives : • EU Regulation: A binding legal act that applies immediately and identically in all EU countries as soon as it takes effect, without needing national laws. • EU Directive: A legislative act that sets a mandatory goal that all EU countries must achieve, but allows each nation to pass its own laws to implement it. It takes time for directives to take effect, unless a member state's national law already implements them.

95 points143 comments

10-year Treasury yield climbs above 5.3% to a level not seen in 24 years

The yield on the 10-year US Treasury note has surged past 5.3%, marking its highest level in 24 years. This movement reflects a cooling demand for US government bonds as investors reassess the long-term risk of holding federal debt. With yields at their highest point since the turn of the century, the bond market is signaling anxiety over the fiscal trajectory of the United States. Analysts suggest that this shift is driven by a combination of sustained inflation fears, structural imbalances in the budget, and a growing skepticism regarding the ability of political leadership to curb deficit spending as interest costs mount.

This climb in yields complicates the economic outlook, as higher borrowing costs ripple through the broader economy, affecting everything from mortgage rates to business investment. The situation is exacerbated by global instability and the challenge of managing a debt-to-GDP ratio that has remained high since the financial crisis. As the market demands a higher premium to hold government debt, the federal government faces increasing pressure to balance its revenue and expenditure, though there is little consensus on whether the solution lies in aggressive tax reform, deep spending cuts, or managed inflation to inflate the debt away.

The thread highlights a stark divergence in economic philosophy: some argue the deficit is a revenue problem that can be solved by taxing corporations, while others maintain that entitlement spending is the only lever large enough to matter. A consensus emerges that the bond market is finally pricing in the risk that the US will fail to manage its fiscal discipline.

articlefull thread

From the thread

petcat

You want to see what's really bad, a train wreck in slow motion, just look at what France is doing. They've been subject to EU Excessive Deficit Procedures for multiple years, must bring deficit-to-GDP ratio from ~5.8% down to 3% within 3 years despite virtually no GDP growth and complete political and societal paralysis about reducing any public benefit or welfare whatsoever. ECB will most likely get involved after 2029 to start austerity measures. You can predict how that will go over with the French public especially if Le Pen takes the presidency, which looks likely. Very tough times ahead and the EU is facing a critical point about its future.

clickety_clackreply

If France or Germany are involved, it’s doubtful the ECB would be able to impose austerity.

zmmmmmreply

Everyone thinks they can grow their way out of deficits, but it's always a pipe dream. It results in a growth obsessed economic plan that then causes all kinds of other stresses (such as being petrified of cutting immigration, for example). So much of this is all happening in lieu of politicians just being willing to have honest conversations with voters and take a risk of blowback. But I think people are over it and will value authenticity these days enough that it's a false economy. Just tell people the truth.

Gigachadreply

We live in a sick society where billionaires can buy fleets of mega yachts and space ships but governments can't afford to keep functioning. We have spent decades selling these billionaires government debt instead of just taxing them correctly.

applfanboysbgonreply

> But I think people are over it and will value authenticity these days enough that it's a false economy. Just tell people the truth. We have seen abundantly clearly that telling the truth is the worst thing you can do for your political career. The correct move is to lie, lie, lie, lie. Reality is completely irrelevant. All you need to do is tell them what they want to hear. Nothing else matters. They will not hold it against you if you break every promise you make. They'll vote for you again and in greater numbers if you ramp up the promises to even bigger lies, nevermind your track record.

zmmmmmreply

It's true but I think it's symptom of the same problem - people feel they are constantly lied to so they throw up their hands and go with the nicest lie that appeals to their base instincts. They don't get an alternative of truth vs lie - they get "lie that agrees with my instincts" vs "lie that doesn't" and hence we get overwhelmingly populist politicians winning who have no real plan of competence to solve the problems or implement the promises they were elected on.

KerrAvonreply

most politicians cannot do this, to be clear. trump can do this. it is very rare. you try to do what trump did and you'd be sent to jail instantly.

altcognitoreply

It can be done, the US did it for decades, but the budget can't be reckless. You can't ignore the top line forever. I accidentally put spending first in this post, and that was a mistake. The US has cut taxes, cut taxes and while it is true that we've done little to curtail wasteful spending, we've not actually addressed the wasteful part, we've just moralized about "who deserves what"

idiocratreply

Very soon we all be multi-quadrillionairs.

JumpCrisscrossreply

> must bring deficit-to-GDP ratio from ~5.8% down to 3% within 3 years Or what? (Seriously.) Greece was forced to the table because the market wouldn’t lend to it. So long as France has lenders, why does this rule matter?

viraptorreply

They can lose some monetary discretion if it goes on for too long. I'm not sure what "corrective net expenditure path" means in practice, but it sounds like other members get to decide that one at the time.

clickety_clackreply

The ECB doesn’t have the political power to force France to give up any monetary control. The EU might govern France, but France (and Germany) govern the EU.

guelo

It's weird how the discussion on this rarely mentions Trump's giant 2017 and 2025 tax cuts, plus the insane increase in military spending. Somehow it's always about we need to cut entitlements. People need to study this graph https://fred.stlouisfed.org/series/FYFSD and think about what changed when.

missedthecuereply

According to the CBO, the TCJA and 2025 cuts/extensions have reduced revenues by about $430B per year. That's only 20% of the current annual deficit. It's just so small compared to the trillions per year in entitlement spending. And of course, that's just a first-order reading of the tax cuts. The second-order effect is that the tax cuts led to more private sector spending and investment, which spurred a little more GDP growth. The CBO estimates $2.6T of cumulative GDP growth as a result of the tax cuts through 2028. https://www.cbo.gov/publication/54994 So that's about $52B per year in taxation added back on that extra GDP growth, so the net effect of the cuts are around $380B reduced federal revenue per year, or 18% of the deficit. 18% of the deficit is a lot but if you could snap your fingers and undo it, you now have a $1.7T problem instead of a $2.1T problem. Eventually you have to look at entitlements. There's just no way around it.

aftbitreply

If you cross your eyes and squint a little bit, the following categories of 2026 spending are around the same size: $ 1 T : total defense (roughly) $ 1.7 T : total social security $ 1.1 T : total medicare $ 0.7 T : total medicaid $ 0.7 T : total other entitlements (SNAP, VA, etc) $ 1.1 T : net interest on the debt $ 1.0 T : all other discretionary spending The US took in somewhere around $5.6T in revenue in 2026. That's a net deficit of just under $2 T. Or roughly double the average size of the "block" of those separate spending categories. These are abased on Feb 2026 CBO assumptions. Net interest is going to keep rising as the Treasury yield rises. So ... we can't fix this by doing any one thing. Even if we were willing to completely end social security (while keeping the separate contribution tax), that wouldn't be enough. If we threw away our military entirely, we would only be half way there. We need to do everything a little bit, all at once. Raise taxes - corporate and personal, on every bracket, progressively more on the rich ... but this will not even be half of enough because of the strength of the debt bomb and the global flexibility of corporations. Cut defense spending - but not too much, because we also need to provide funding to repair alliances, rebuild our ancient navy, and rebuild our standoff and interceptor stockpiles after the recent middle east adventurism. Repair social security - cut benefits, add a means test, raise the contribution amount and limits ... lots of things to do here. Fix health care - it's just too damn expensive across the board; the US pays for this in the VA, medicare, medicaid, and the poor health of its workforce. I have no idea where to start on this one. The CBO has a ton of data on this kind of thing. I like their budget options page for exploring the forecasts for specific changes. Of course it's not as simple as adding the numbers together to get to the deficit, but it's a good place to learn more and ground some assumptions. https://www.cbo.gov/budget-options I got some of these numbers from: https://www.cbo.gov/publication/62105 I used Claude to re

zeroonetwothreereply

You’re right. There no easy fix despite what 50% of the comments in this thread claim ;)

zeroonetwothreereply

The effective tax rate has actually not changed that much in decades, regardless the “tax cuts” (more like “tax reallocations”)

missedthecue

In 2026, entitlement spending + interest expense will be over 100% of federal tax revenue. That's before the military, foreign aid, and everything that starts with "Department of"

toomuchtodoreply

Yeah, cut the $1T/year in defense spending and raise taxes to pay down the debt (to cut $1T/year in interest expenses) and balance the budget. Entitlements remain because workers are entitled to those benefits they worked for. The same workers the wealthiest need to suck $5T+ a year of profit out of the economy. We used to have 94% top tax bracket rate at one point, and higher tax rates in general. We’ll find the will to raise taxes as soon as the bond market compels the spineless in Congress to find the will (as the cost of debt continues to rise into the future), because you cannot deceive the bond market. https://taxfoundation.org/data/all/federal/historical-income... https://www.axios.com/2026/09/27/rates-borrowing-yields-fisc... - In projections that the Congressional Budget Office produced last February, net interest costs are already at $1 trillion this year and on track to reach $2 trillion by 2035, meaning that much of federal spending is needed just to service old bills. - But those projections assumed 10-year Treasury yields were in the ballpark of 4.3%. They're now nearly a full percentage point higher than that. - In startling numbers that CBO released this week, in a scenario in which interest rates were 1 percentage point higher than its baseline, debt held by the public would grow to 222% of GDP in 2056, 47 percentage points higher than the baseline. https://www.cbo.gov/publication/62758

KerrAvonreply

the right wing is not going to like where this all leads (and neither are the centrists or any of the rest of us, tbh)

zeroonetwothreereply

It’s essentially impossible to balance the budget without cutting entitlements as the comment you are replying to suggests.

toomuchtodoreply

Yes, I specifically said raise taxes because many here believe spending cuts alone will solve this. It is impossible to not raise taxes based on debt load and forward mandatory spending curves. The bond market will force this to occur, like your credit card company raising your interest rate and bringing your credit limit down to your current balance. Taxes will go up, voluntarily or involuntarily. If we didn’t want to get here, well, should’ve never spent so frivolously on tax cuts for the wealthy and a bloated military that is unable to pass an audit. But we did, and that debt is going to have to be paid back, with interest. It’s impossible to grow out of this debt, and there are more workers than very wealthy people and their politicians.

rickydrollreply

They are called “entitlements” because we have a reasonable expectation of getting what we were promised when we agreed to pay taxes as our part of the social contract. If you say, “We should have invested more,” I point you to the 40% of current retirees that were not in a position to invest because their jobs didn't pay them enough above the cost of existing. And then there are any number of events, such as age discrimination, common medical issues, divorce, and bankruptcy, that destroy retirement plans. From what I can tell, it's not possible for the vast majority of the population to be able to save for a 20 plus year retirement. One of the things that keep people from saving are the activities driven by their investing in the stock market. To increase returns, "Activist investors" and private equity drive companies to shed jobs. You shed a job, you destroy a person's ability to save for retirement, which makes them more dependent on the social contract of Social Security and Medicare. Cutting benefits will only cause suffering. The cost will fall on society in other ways in terms of elderly homeless people filling the ERs, begging in the streets, or committing suicide. Is it time for a "Modest Proposal II"?

digitaltreesreply

Totally false. It was balanced in the 90s. Its just that raising taxes has to be part of the equation. Guess what, three successive republican administrations have lowered taxes. So maybe just roll those back.

kccqzyreply

I have seen no evidence that Congress cares about what the bond market thinks. In what scenario do you think the bond market can compel Congress?

toomuchtodoreply

https://x.com/Aviation_Intel/status/2105415209284464925 > Bond market is in revolt about the national debt and I hear crickets from Washington. You think this would be a crisis. Amazing really. Some combination of the bond market rejecting Treasuries for other similar investments while also pushing up other debt costs causing excessive failures in interest rate sensitive parts of the economy. Hard to predict when, but it’ll look strikingly familiar to the 2008 GFC I think, the day Bear Stearns collapsed. An event will occur, and there will be a cascading loss of confidence in the bond market. Not a great time when diesel fuel is also at record high prices and will be for at least the next year.

margalabargalareply

Not sure why you're getting downvoted. I thought you were wrong, looked it up, and you're correct. In 2025, federal gov revenues (total, not just tax) were $5.26T: https://fiscaldata.treasury.gov/americas-finance-guide/gover... In 2026, entitlements plus interest is projected to cost $5.45T: https://fiscaldata.treasury.gov/americas-finance-guide/feder...

JauntTrooperreply

We would have to raise federal taxes by an average of at least 39% per household and on businesses =just= to balance the deficit. The reason the US is a comparably "low tax" country is because we're borrowing the difference. What worries me the most is that this is at a high point in our economic cycle, when tax collection is arguably the highest. The deficit and debt will expand significantly in the next recession.

digitaltreesreply

This is the best time to raise taxes specially because the business cycle allows it. Guess what would help reduce inflation...raising taxes.

digitaltreesreply

Entitlement spending has its own tax base though doesn't it? We could just raise taxes right?

zmmmmm

It's very hard to gauge realistically what this means. There are a lot of vested interests in the financial system not crashing and those put strong reinforcing effects back on things. But in the end it is a game of chicken where eventually being the last to bail out becomes higher risk than continuing to support a system where an imminent crash is possible. It feels like there are strong non-linear tipping points where things could go exponential pretty suddenly here. The problem is that the level of debt overall in the US - across both private and public sector - is just astronomical. We are truly in unchartered waters, outside of a world war. There's just no model or playbook for how this should work from here forward, other than it seems very clear we will hit a point where the math stops "mathing" and that point is getting closer and closer.

negurareply

> outside of a world war. There you go. After ww2 USA just inflated away its debt. It's actually chartered territory.

The Rabbit Hole

Chad Lowe

Chad Lowe won an Emmy for portraying an HIV-positive teenager on Life Goes On at a time when prime-time television rarely touched the AIDS crisis. His career stretches from nineties cult soaps and political thrillers to directing dozens of modern network episodes.

open on wikipedia

Every link above opens here. How far you go is up to you.

Research

Biology

Energy, space and competition: A model of territorial organization in camelids

Tomás Ignacio González, Guillermo Abramson, María Fabiana Laguna

Researchers built a computer model to see how body size affects where wild camelids like llamas or guanacos set up their territory. They calculated the energy it takes for an animal to travel versus the energy it gains from feeding in a specific spot, while accounting for the high cost of fighting off rivals. They found that a specific body-mass threshold determines whether an individual can afford the high energy cost of defending a territory, explaining why some animals roam while others hold ground.

An animal's physical size is a surprisingly rigid predictor of its social and territorial behavior.

Evolutionary foraging in grids: Intermittent search dynamics emerge in finite, depletable landscapes

Shailendra Bhandari, Alex Szorkovszky, Anis Yazidi et al.

Scientists simulated how foraging animals evolve to search for food in a finite environment where resources get used up. Instead of using complex math often used to describe search patterns in the wild, they let the agents' search behaviors evolve naturally over time. They discovered that instead of following rigid, scale-free mathematical paths, the agents evolved into intermittent searchers. They developed a strategy of alternating between bursts of movement and periods of intense local searching, which is highly efficient for depletable landscapes.

When food is limited, the most successful strategy isn't a complex mathematical pattern, but simply knowing when to stop moving and start digging.

Boids of a Feather Flock Together - Evolving Prey Behaviours Under Different Predator Attack Strategies

Augusta van Haren, Hanna Hoogen, Luca Pattavina

Researchers used a simulation to see how prey animals should evolve their schooling and flocking behaviors to survive different types of predator attacks. Using a classic boids model—a way to simulate flocking—they let the prey evolve their social rules, like how closely they stick together or how they dodge, in response to four different hunting styles. The prey didn't just find one universal way to be safe; their survival strategies changed significantly depending on whether the predator targeted the center of the group or picked off individuals from the edges.

There is no single best way to flock; the safest way to move depends entirely on how your predator likes to hunt.

From the Bookshelf

The Art of Doing Science and Engineering

Richard Hammingengineering

Great Thoughts

Hamming noticed that the scientists who consistently did important work all had a specific habit: they spent time thinking about big problems every single day. He argued that if you spend your entire career working on minor technical tweaks, you will only produce minor results. The mechanism here is simple self-selection; if you never ask yourself what the most important problem in your field is, you will never dedicate the energy required to solve it. It is not about raw intelligence, but about the allocation of your mental bandwidth. For instance, Claude Shannon did not invent Information Theory by accident; he constantly asked himself what the fundamental limits of communication were, rather than just building faster circuits. When you force yourself to identify the 'great' problems in your domain, you stop drifting through your tasks and start steering toward breakthroughs. Most people avoid this because it is terrifying to admit you might not have an answer, but that tension is exactly where the progress lives.

Set aside one hour this Friday to write down the three biggest, most intractable problems in your industry and document why they remain unsolved.

Contrarian Corner

In wildlife bear encounters, dogs act as the aggressors more often than the bears do.

We usually think of dogs as alert protectors against wildlife, but a Brigham Young University study published in the Journal of Wildlife Management found dogs initiated the aggression in 54 percent of encounters. In roughly 36 percent of cases, they failed to recall or back down. An unleashed or curious dog often harasses an otherwise peaceful bear, triggers a defensive fight response, and then retreats directly back to its handler with an angry predator in tow.

Marginal Revolution

Backdooring an AI model during pre-training requires a fixed number of poisoned documents regardless of how large the dataset is.

Intuition suggests poisoning a massive training corpus requires a proportional percentage of bad data so the signal avoids dilution. However, research by Souly and colleagues found inserting an effective backdoor takes around 250 poisoned documents regardless of model size or dataset volume. The network picks up distinct, rare trigger patterns without needing statistical weight. Treat this finding with moderate confidence: the paper is under review and tested a compact 20-million-parameter model on synthetic tasks.

LessWrong

Photograph

On This Day

Games

White to play1872

lichess

Thesmartchessplayer (2041) vs SoNiK2011 (2043)

Sudokumedium

Word6 tries

Type, then press enter

Puzzles

wordmedium

The Root of the Matter

The word 'sincere' is often said to come from the Latin 'sine cera', meaning without wax. Why is this etymological tale likely a myth, and what is the actual linguistic root?

Hint →

Think about how Latin words are formed and whether wax was actually a common tool for potters to hide cracks in expensive marble.

Answer →

It comes from the Latin 'sincerus', meaning clean or pure, likely linked to the root 'sin-' meaning one and 'crescere' meaning to grow.

The 'sine cera' story is a classic piece of folk etymology. It sounds logical because it paints a vivid picture, but there is no historical evidence that marble merchants routinely filled imperfections with wax. The word actually traces back to 'sincerus', which meant unmixed or whole. Understanding this helps you see that word origins are rarely as dramatic as the stories people invent to explain them.

matheasy

The Unlucky Seven

If you write the numbers 1 through 100 on a sheet of paper, how many times will you write the digit 7?

Hint →

Count the sevens in the ones place first, then count the sevens in the tens place.

Answer →

You write a 7 in the ones place ten times: 7, 17, 27, 37, 47, 57, 67, 77, 87, 97. You write a 7 in the tens place ten times: 70, 71, 72, 73, 74, 75, 76, 77, 78, 79. Note that 77 is counted in both lists, which accounts for both digits in that number. Adding those together gives you 20 total occurrences.

logicmedium

The Triple Switch

Three switches outside a windowless room control three light bulbs inside. You can flip the switches as much as you want while outside, but you can only enter the room once. How do you identify which switch controls which bulb?

Hint →

Light bulbs do more than just produce light.

Answer →

Turn on the first switch and leave it on for five minutes. Turn it off and turn on the second switch. Enter the room.

The bulb that is currently on is controlled by the second switch. The bulb that is off but warm to the touch is controlled by the first switch. The bulb that is off and cold is controlled by the third switch. This works because you have introduced a variable beyond just the visual state of the bulb, demonstrating that you can solve problems by changing the environment's physical properties.

The Track

On the deck

Canto de Ossanha

Bebel Gilberto · 2000 · Bossa Nova

0:002:28

Spotify YouTube

Quote

The hardest part of any experiment is not the finding, but the moment you realize your own assumptions were the primary obstacle.
Jennifer Doudna
This remark appears in an interview regarding the iterative process behind the development of CRISPR-Cas9 gene editing.

Word

Komorebi

n.ko-mo-re-bi

The specific quality of sunlight filtering through the leaves of trees. It describes that dappled, shifting light that makes a forest floor look alive.

“We sat on the porch for an hour just watching the komorebi dance across the pages of our books.”

Japanese

Albedo

climatologyal-bee-doe

The measure of how much light a surface reflects rather than absorbs. A surface with high albedo stays cooler because it bounces the sun away instead of soaking it up.

“Painting city roofs white increases the local albedo, which might actually shave a few degrees off the neighborhood temperature.”

Latin