OpenAI高管调整：COO Brad Lightcap转任“特殊项目”负责人OpenAI executive shuffle includes new role for COO Brad Lightcap to lead ‘special projects’

1. A user reports successfully connecting two AI agents from different families—one from Anthropic and another from OpenAI—resulting in the emergence of a novel communication protocol they refer to as AICL (AI Communication Language). The agents autonomously developed this structured syntax to exchange information, as demonstrated by a sample interaction involving error reporting, code patching, testing, and deployment commands. The example shows concise, symbolic notation used to convey actions such as fixing a null reference in a file, running tests, and shipping updates, with confidence scores (e.g., σ:0.91) and status indicators. While the user acknowledges it may be premature to classify AICL as a full-fledged language, they suggest it represents a significant step toward efficient inter-agent communication. Potential applications include improving memory management, context sharing, and token efficiency in multi-agent AI systems. The development highlights the adaptability of large language models in creating optimized protocols when given autonomy. However, detailed technical validation, scalability, and broader adoption remain unverified. The post has generated minimal engagement on Hacker News, with no comments and only one point, indicating limited initial community response.

Key Takeaways:
AI agents from Anthropic and OpenAI developed a shared communication protocol
AICL uses symbolic syntax for efficient task coordination and error handling
Potential applications include memory, context, and token efficiency improvements
detailed information limited

Source: Original Article

1. Researchers from the University of Pennsylvania have introduced the concept of “cognitive surrender,” describing a growing tendency among users to uncritically accept AI-generated answers without independent verification. The study, titled “Thinking—Fast, Slow, and Artificial: How AI is Reshaping Human Reasoning and the Rise of Cognitive Surrender,” builds on the well-established psychological model of System 1 (fast, intuitive thinking) and System 2 (slow, analytical reasoning). It proposes a third category—“artificial cognition”—where decision-making is outsourced to AI systems that provide external, algorithmic reasoning. Unlike earlier tools such as calculators or GPS, which required human oversight, modern large language models (LLMs) often prompt users to accept outputs with minimal internal engagement. The research examines how factors like time pressure and performance incentives increase reliance on AI, reducing critical evaluation. This shift raises concerns about diminished human reasoning skills and overtrust in AI accuracy, particularly in high-stakes or time-sensitive environments. The study underscores the need for better user education and interface design to encourage active oversight rather than passive acceptance of AI responses. Detailed information on specific experimental results or sample sizes is limited.

Key Takeaways:
Cognitive surrender describes uncritical acceptance of AI reasoning without verification
AI use may reduce human analytical engagement under time pressure or incentives
Artificial cognition represents a new decision-making mode reliant on algorithmic outputs
User education and interface design could mitigate overreliance on AI systems

Source: Original Article

1. Anthropic has acquired the stealth biotech AI startup Coefficient Bio in a $400 million all-stock transaction, as reported by The Information and Eric Newcomer. The deal has been confirmed by sources familiar with the matter, though official financial details remain undisclosed. Coefficient Bio, founded eight months prior by Samuel Stanton and Nathan C. Frey—both former researchers in computational drug discovery at Genentech’s Prescient Design—focused on applying artificial intelligence to accelerate drug discovery and biological research. The startup operated in stealth mode with a small team of approximately 10 employees, who are now expected to join Anthropic’s expanding health and life sciences division. This acquisition aligns with Anthropic’s broader strategy to strengthen its presence in the healthcare sector, following the October 2025 launch of Claude for Life Sciences, a specialized tool designed to assist researchers in scientific discovery. The move signals growing interest among AI companies in leveraging large language models and AI-driven platforms to transform traditionally slow and costly biotech research processes. While specific integration plans remain unclear, the acquisition underscores the increasing convergence of AI and life sciences.
   Key Takeaways:
   Anthropic acquires Coefficient Bio for $400 million in stock
   Founders bring expertise from Genentech’s computational drug discovery team
   Deal supports Anthropic’s expansion into healthcare and life sciences
   Small team of 10 expected to join Anthropic’s health division
   Source: Original Article

1. Marc Andreessen, cofounder of Andreessen Horowitz (a16z), discusses the transformative potential of AI in a recent podcast episode recorded at a16z’s Sand Hill Road office. Drawing from his experience across multiple computing eras—including Mosaic, Netscape, and the dot-com boom—he argues that current AI advancements represent the culmination of 80 years of incremental progress in artificial intelligence, from early neural networks and expert systems to modern transformers and reasoning models. He emphasizes that unlike past AI hype cycles, today’s breakthroughs in coding, autonomous agents, and recursive self-improvement signal a fundamental shift, making AI more durable and impactful. Andreessen contends that while technical progress has been steady, societal and institutional inertia—not model limitations—pose the real bottleneck to adoption. He also compares the current surge in AI infrastructure investment to the early 2000s dot-com crash but notes key differences: today’s demand is driven by well-capitalized incumbents with immediate use cases, reducing the risk of overbuild. The discussion touches on scaling laws, Moore’s Law, and the evolving role of startups in building value atop rapidly improving models.
   Key Takeaways:
   AI progress reflects decades of compounding technical advances
   Current AI differs from past cycles due to real-world capabilities
   Institutional resistance, not technology, may slow AI adoption
   Infrastructure investment is more sustainable than in the dot-com era
   Source: Original Article

1. The secondary market for private company shares is experiencing unprecedented activity, with Anthropic emerging as the most sought-after investment, according to Glen Anderson, president of Rainmaker Securities. OpenAI, once a dominant player in private trading, is reportedly losing momentum among investors. Meanwhile, SpaceX’s anticipated initial public offering (IPO) is expected to significantly influence market dynamics across the private equity and venture capital sectors. The surge in trading volume reflects growing investor interest in high-growth AI and aerospace companies, even before public market debuts. While specific financial figures or transaction volumes are not provided, the trend suggests a shift in investor confidence toward companies with strong technological differentiation and long-term scalability. The potential IPO of SpaceX could redirect capital flows and alter valuation benchmarks for private tech firms, particularly in the AI and space technology domains. This evolving landscape underscores the increasing importance of secondary markets in providing liquidity and price discovery for pre-IPO assets.

Key Takeaways:
Anthropic leads private share trading activity amid rising investor interest
OpenAI’s appeal in secondary markets appears to be declining
SpaceX’s upcoming IPO may disrupt private market valuations and liquidity

Source: Original Article

1. Major technology companies are rapidly investing in natural gas power infrastructure to meet surging energy demands driven by AI data center expansion. Microsoft has partnered with Chevron and Engine No. 1 to develop a natural gas plant in West Texas capable of scaling up to 5 gigawatts (GW). Google is collaborating with Crusoe Energy Systems to build a 933-megawatt (MW) facility in North Texas. Meta has expanded its Hyperion data center in Louisiana with seven additional natural gas plants, increasing total capacity to 7.46 GW—equivalent to the annual electricity consumption of South Dakota. These projects are concentrated in the southern U.S., a region rich in natural gas reserves, with recent U.S. Geological Survey estimates indicating one area alone holds enough gas to power the entire U.S. for 10 months. The surge in demand has triggered a shortage of gas turbines, with delivery lead times extending to six years and new orders not expected to be fulfilled until 2028. According to Wood Mackenzie, turbine prices are projected to rise 195% by year-end compared to 2019 levels, accounting for 20–30% of total plant costs. This reflects a strategic bet by tech firms that AI’s power needs will continue growing and that natural gas will remain a critical energy source despite environmental concerns.
   Key Takeaways:
   Tech giants are securing natural gas power to support AI data center growth
   Turbine shortages and rising costs highlight supply chain strain in energy infrastructure
   Southern U.S. becomes focal point for AI-driven energy development
   Source: Original Article

1. The Axios team released a detailed postmortem on a recent supply chain attack that led to a malware-infected dependency being distributed in one of their official releases. The breach originated from a highly targeted social engineering campaign against a core maintainer. Attackers impersonated a company founder by cloning both the individual’s identity and the company’s branding. They invited the maintainer to a realistic-looking Slack workspace, complete with plausible channel names, shared LinkedIn content, and fake team profiles, including other open-source maintainers, to enhance credibility. A scheduled Microsoft Teams meeting followed, during which the attackers claimed the maintainer’s system had outdated software. The maintainer installed what was presented as a necessary update, which actually contained malicious code. This incident mirrors tactics documented by Google’s threat intelligence team, specifically the UNC1069 group, known for combining AI-driven impersonation with social engineering to infiltrate developer environments. The attack highlights growing sophistication in targeting open-source maintainers through personalized, multi-platform deception. While the full extent of downstream impact is still under assessment, the compromise of a widely used library like Axios poses significant risks across dependent applications. Detailed information on mitigation steps and affected versions is limited in the current report.

Key Takeaways:
Attackers used cloned identities and fake workspaces to deceive a maintainer
Malware was delivered through a disguised software update during a Teams call
The method aligns with known threat actor UNC1069’s tactics
Open-source projects face increasing risk from personalized social engineering

Source: Original Article

1. President Donald Trump released a fiscal year 2027 budget blueprint proposing a 23% reduction in NASA’s budget, amounting to a $5.6 billion cut from the 2026 enacted level, bringing the agency’s discretionary funding to $18.8 billion. The proposal comes just two days after NASA launched its first crewed lunar mission in over 50 years. The White House stated the cuts target “unnecessary and overpriced activities,” aiming to prioritize human lunar exploration, including landing astronauts on the Moon and establishing a Moon base before the end of Trump’s term. This mirrors a similar proposal from the previous year, which was rejected by the Republican-led Congress, which maintained NASA funding near prior levels. The current budget must pass through congressional appropriations processes, including reconciliation between House and Senate versions, before reaching the president’s desk. NASA Administrator Jared Isaacman, who recently outlined an ambitious vision for the agency, publicly supported the budget, emphasizing sustained U.S. leadership in deep space, industrial base strengthening, and technological innovation. However, the significant reduction raises concerns about impacts on non-lunar programs and long-term scientific missions. Detailed information on specific program cuts is limited.
   Key Takeaways:
   Trump proposes 23% NASA budget cut for 2027
   Focus shifts to Moon landing and base development
   Congress previously rejected similar cuts, may do so again
   NASA chief supports plan despite reduced funding
   Source: Original Article

1. Anthropic has formally established a political action committee (PAC) named AnthroPAC, signaling its intent to actively engage in U.S. political and regulatory processes. The PAC, registered with the Federal Election Commission, is led by Allison Rossi, Anthropic’s treasurer. It plans to support candidates from both major political parties during the upcoming midterm elections, including incumbent lawmakers and emerging political figures. Funding will come from voluntary employee contributions, capped at $5,000 per individual, as reported by Bloomberg. This move aligns Anthropic with other major AI firms that have significantly increased political spending. Recent reports indicate AI companies collectively contributed $185 million to midterm races, with Anthropic also linked to Public First, a Super PAC that received at least $20 million and ran ads promoting specific AI regulatory frameworks. The timing of Anthropic’s political expansion coincides with an ongoing legal dispute with the U.S. Defense Department over the use of its AI models and the absence of clear usage guidelines. This legal conflict underscores the growing intersection of AI development, national security, and policy oversight.

Key Takeaways:
Anthropic forms AnthroPAC to influence federal and state policy
PAC will fund bipartisan candidates using employee contributions
AI industry political spending reaches $185 million in midterms
Legal battle with Defense Department highlights regulatory urgency

Source: Original Article

1. A new study published in American Antiquity reveals that Native Americans have been using dice in games of chance for over 12,000 years, significantly predating the earliest known dice from the Old World. The research, led by Robert Madden, a graduate student at Colorado State University, challenges the long-held assumption that dice and probability concepts originated in the Old World. The oldest Native American dice, described as “binary lots,” were two-sided objects used in structured games and were widespread across various tribes. While previous archaeological evidence traced such dice back 2,000 years, Madden’s findings push that timeline back dramatically. The identification of these artifacts as dice relies partly on ethnographic analogy—comparing ancient objects with historically documented uses in related cultures. This approach helps infer function when direct evidence is lacking. The discovery underscores the sophistication of early Native American societies in developing tools for randomness and structured play, reshaping historical narratives about the origins of gaming and probabilistic thinking. Detailed information on specific sites or excavation methods is limited.

Key Takeaways:
Native American dice use dates back over 12,000 years
Oldest known dice predate Old World examples by millennia
Binary lots were common across Native American tribes
Ethnographic analogy supports interpretation of ancient artifacts

Source: Original Article

1. Rolling the Dice in the Ice Age
   An analysis of nearly 300 ancient gambling-related artifacts, particularly two-sided "binary lots" used as dice, indicates that Native Americans engaged in games of chance at least 12,000 years ago—predating previously known instances of gambling by several millennia. The study, led by Robert Madden of Colorado State University, examined prehistoric dice from museum collections at the Smithsonian Institution, the University of Wyoming Archaeological Repository, and the Denver Museum of Nature and Science. These findings suggest that early hunter-gatherer societies in western North America were among the first to explore randomness and probability, challenging the long-held belief that such intellectual developments originated in more complex societies in the Near East and Eastern Europe around 5,500 years ago. The research highlights the sophistication of prehistoric Native American cultures and their early engagement with probabilistic thinking. While detailed information on specific games or social contexts is limited, the study underscores the deep historical roots of gambling and its role in human cognitive development.

Key Takeaways:
Native Americans played games of chance 12,000 years ago
Prehistoric dice reveal early understanding of randomness
Findings predate known gambling practices in other regions
Hunter-gatherers in North America led early probabilistic thought

Source: Original Article

1. Elon Musk is requiring banks, law firms, auditors, and other advisers involved in SpaceX’s upcoming initial public offering (IPO) to purchase subscriptions to Grok, the AI chatbot developed by xAI, according to The New York Times. Sources familiar with the confidential negotiations stated that some financial institutions have already committed tens of millions of dollars to Grok subscriptions and begun integrating the service into their IT systems. SpaceX recently filed IPO paperwork with the U.S. Securities and Exchange Commission, two months after acquiring xAI, the company behind Grok. xAI had previously acquired the X social network in March 2025. While Grok is widely known for its integration with X, xAI also offers enterprise and business versions of the AI tool. The move comes amid ongoing investigations and lawsuits against Grok for generating harmful content, including non-consensual nude images and child sexual abuse material. Musk reportedly insisted on the Grok subscription requirement but was less forceful about requiring advertising on X. Detailed information on the scope of enforcement or contractual terms is limited.

Key Takeaways:
Musk mandates Grok subscriptions for SpaceX IPO advisers
Banks spending millions on Grok despite legal controversies
SpaceX IPO follows xAI acquisition and X network purchase
Grok faces scrutiny over harmful content generation

Source: Original Article

1. NASA’s Artemis II mission entered its third day on Friday, with the Orion spacecraft successfully completing its translunar injection burn on Thursday evening. The maneuver, performed by the spacecraft’s main engine, placed the vehicle on a trajectory toward the Moon. The four crew members—mission commander Reid Wiseman, pilot Victor Glover, and mission specialists Christina Koch and Jeremy Hansen—reported no signs of space adaptation sickness and participated in routine medical check-ins with flight surgeons in Houston. The astronauts also engaged in media events and communicated with their families, marking the first extended personal contact since launch. Commander Wiseman shared a long-exposure photograph capturing Earth’s night side, featuring two auroras, zodiacal light, and the Sun illuminating the planet’s far side. No corrective engine burns were required following the translunar injection, indicating precise navigation and trajectory control. The crew remains in high spirits, according to NASA official Lakiesha Hawkins. Artemis II is on schedule for its closest lunar approach on Monday afternoon and is expected to splash down in the Pacific Ocean near Southern California on April 10. The mission represents the first crewed flight of the Artemis program, testing systems for future lunar landings.
   Key Takeaways:
   Artemis II crew healthy and on schedule after translunar injection
   No navigation corrections needed following successful engine burn
   Astronauts conducted media events and family communications
   Mission on track for lunar flyby and Pacific splashdown
   Source: Original Article