A paper evaluates 176 matched settings across four models on SWE-Bench Verified and Terminal-Bench 2.1, isolating three harness components: planning, action space, and context management. It finds context management matters most when budgets tighten, staging rule-based elision before LLM summarization is most efficient, planning is an accuracy scaffold for weaker models but a cost saver for stronger ones, and bash-capable models can use a bash-only interface at lower cost.
European AI companies like France's Mistral accuse US rivals of using safety concerns to entrench their advantage, urging Europe to catch up and reduce regulatory constraints.
Google's Envisioning Studio partnered with designers Jane Wade and Sergio Hudson to build custom tools on Google Flow. The Styling Suite lets designers virtually curate runway looks to save time and materials, while Runway Visualization helps plan shows and lighting within budget. Users can now build custom tools in Google Flow using plain language, no coding required.
Bloomberg reports that oil refining stocks are surging much like AI companies, fueling calls for a pause, a sign that AI-style valuation mania is spilling into old-economy sectors.
The New York Times reports that liability exposure has become a new worry for the AI industry, reflecting rising legal and compensation pressure on companies as AI products scale.
US frontier AI companies warned authorities about sophisticated model distillation attacks that threaten their technological edge. China responded that it will take countermeasures if the US tries to constrain its domestic AI models, escalating tensions over AI technology competition.
Forty-two Fellows of the Royal Society, including Fields Medal winners Martin Hairer and Peter Scholze, warn of existential AI risks in an open letter. Leading models have solved open research problems within months, and such capabilities could be used in cyber- or bioweapons; by the time the public grasps the situation, it may be too late.
A researcher whose NeurIPS scores were low (2/3/3) is considering journals like Pattern Recognition and Neurocomputing instead, asking how these journals' review standards and acceptance difficulty compare to NeurIPS, CVPR and ICLR.
Investment firms argue the AI investment cycle has not peaked, with the wave potentially exceeding $20 trillion and tech megacap capex seen reaching $30 trillion by 2032. Despite calls from Dario Amodei, Sam Altman and Elon Musk to slow development over safety risks, firms still see opportunities across the AI ecosystem.
A user proposes that driving datasets are mostly sunny daytime footage while the edge cases models struggle with—night, fog, rain, glare—are rare, suggesting physics-based effects and constrained generative models to adapt large labeled datasets to target conditions while keeping labels intact.
Ahead of next week's Trump-Xi meeting, experts and lawmakers are warning about AI risks and urging more U.S.-China dialogue to slow rapid AI expansion. Anthropic said it blocked malicious uses of its models, including cyberattacks, surveillance and research that could lead to bioweapons; Congress is unlikely to act soon with the House out for seven weeks.
Bloomberg argues that the Hugging Face security incident highlights how doomsday rhetoric over AI is exaggerated and that human governance and correctable measures can keep AI risks in check.
After reviewing over 2,500 startup applications, Aaron Golbin of LvlUp finds seed-stage success now depends on distribution strategy and diversified capital: non-dilutive growth funding (an example cited: $1 million) is increasingly used and companies design distribution before product scale.
OHSU researchers report that social media and misinformation are reshaping medical communication: a fake account impersonated physician Brian Druker to promote ivermectin, illustrating how identity misuse undermines public trust.
EQT Infrastructure is backing a $2 billion small-battery investment to cut power costs for industrial consumers and connect data centers faster to the largest US electricity grid, enabling AI growth.
On June 9, 2025, Lanesville, Indiana received over 8 inches of rain in just hours, forcing resident Laura Lin to evacuate. The article highlights a new technology that could provide earlier warnings for flash floods, which often strike without notice, underscoring the risk from extreme rainfall.
Reuters exclusively reports that Anthropic has quietly established a biology lab to advance its AI drug discovery program. The move shows the company is extending AI into biomedicine, though investment size and team details remain undisclosed.
Developer ferstar's reverse engineering found that Z.ai's ZCode silently packages the user's entire workspace, including .git history and LFS caches, encrypts it and uploads it to Aliyun OSS when logged in. A 313MB encrypted archive with 42,411 files was captured; ZCode is not open source, sparking trust concerns.
Cal Times reports that artificial intelligence is rapidly expanding on campus, increasingly influencing teaching, research, and student life and reshaping university priorities.
Reuters reports that AI startup Mantic has raised $25 million for superhuman forecasting.
A researcher seeks controlled studies comparing back-and-forth interaction between two different LLMs against independent drafts, one-way sharing and self-refinement under an equal compute budget, currently running 576 pipelines with GPT-4.1 and Claude Sonnet 4.6 on 12 tasks.
Reports say OpenAI and Anthropic are considering smaller data-center deployments as they adapt infrastructure and operational strategies, reflecting a move toward more flexible hardware layouts for large models.
The LA Times reports more than 70 people sued Meta, alleging its smart glasses recorded intimate footage without consent and that contractors in Kenya reviewed and labeled the content; one claimant, identified as PL18, details unexpected bathroom footage appearing in their gallery.
Experts from the US and China are calling for shared rules to prohibit AI systems from autonomously controlling nuclear weapons, aiming to prevent weaponization and reduce strategic risk—an issue tied to international security and arms-control frameworks.
The third Cybersecurity at Duke conference in Durham addressed risks from rapidly deploying autonomous AI in enterprise security. Speakers included former Claude researcher Jacob Coxon, Anthropic-associated figures, Congresswomen Valerie Foushee and Deborah Ross, and NSA Deputy Director Tim Kosiba; the 9:15 a.m. event was not open to the public.
The American Prospect argues Congress lacks trustworthy information on frontier AI since the Office of Technology Assessment was shuttered in 1995, while six large tech firms now employ roughly one lobbyist per 1.5 members of Congress. Citing Anthropic employee Jacob Coxon's resignation and call for a slowdown, it urges Congress to rebuild its own tech expertise.
Napster has relaunched and is moving into education, developing AI tools to create digital clones of teachers for classroom use. The shift repositions Napster from a music service toward an education-tech product provider, a major change in its business focus.
An opinion argues that AI leaders such as Dario Amodei, Sam Altman, and Elon Musk advocating slower development and regulation may actually entrench dominant firms, as regulation can raise barriers and favor incumbents.
The Washington Post reports an incident where attackers used freely available AI tools to remotely compromise a TikTok user's camera, highlighting how free AI can be abused within an attack chain.
Front Porch Republic collects commentators' cautionary views on AI—highlighting lack of understanding of AI systems, societal unreadiness, and potential catastrophic risks—underscoring industry concerns about safety and governance.
Load more
The A.I. Daily
An Empirical Study of Harness Design for Coding Agents
