A user reports achieving performance improvements across all benchmarks and securing 1st place on the HuggingFace Open LLM Leaderboard by duplicating intermediate layers of Qwen2-72B without modifying the weights.
OpenAI will reportedly forgo plans to expand the Stargate data center in Abilene, Texas, with Oracle in order to secure higher-performance Nvidia GPUs.
Mozilla announced that Anthropic's red team used Claude to identify security vulnerabilities in Firefox and worked toward their remediation.
OpenAI CEO Sam Altman stated during a Q&A regarding a new contract with the U.S. Department of Defense that he had considered that the construction of AGI should be a government project.
Jensen Huang stated that Nvidia's investments will end with the current round, as investment opportunities will close if both companies go public later this year.
Joel Gavaras has filed the first wrongful death lawsuit in the U.S. against a tech giant, alleging that Google's AI tool, Gemini, fueled his son's delusions and led to his suicide.
OpenAI CEO Sam Altman announced that a new contract with the Department of Defense maintains safety principles prohibiting domestic mass surveillance. However, sources point out that the inclusion of the phrase any lawful use could enable mass surveillance based on past legal interpretations.
Niklas Gustafsson, a Hacker News poster, reports that he has constructed a Transformer model capable of learning 10-digit addition with an extremely small number of parameters.
President Trump has reportedly instructed the Department of Defense to designate AI developer Anthropic as a supply chain risk.
OpenAI has announced that it raised $110 billion based on a pre-money valuation of $730 billion.
Hardware design platform Flux has secured $37M in new funding to enhance its AI agent capabilities and core platform.
unf, a local-first automatic file snapshot tool developed following an experience where an AI agent overwrote working files, has been released. It records files—including those not yet committed to Git—upon every save, allowing users to roll back to any point in time.
Google has announced the launch of SynthID, a tool for embedding watermarks into AI-generated images, audio, text, and video, now supporting text generated via Gemini apps and web experiences.
Google DeepMind has announced Nano Banana 2, a new image model that combines the advanced capabilities of Nano Banana Pro with the speed of Gemini Flash. It will be available across existing services, including the Gemini app and Google Search.
A developer has released Clocksimulator.com, a minimalist analog clock web app free of ads and tracking. It features OLED burn-in prevention and offline availability as a PWA.
Pi is a minimal AI agent harness that users can freely customize with extensions and skills.
Security researchers investigating infrastructure related to OpenAI, the US government, and Persona have discovered public information indicating the existence of a surveillance database.
xAI has signed an agreement to make Grok available for use within US military classified systems. Amid tensions with Anthropic, the path is opening for other AI models to enter the space.
xAI's Grok will be available in the US Department of Defense's classified systems. The trend of introducing multiple models into top-secret environments, previously dominated by Anthropic's Claude, is accelerating.
At an AI summit in India, CEO Sam Altman claimed that the energy used for ChatGPT's responses is equivalent to or more efficient than that of humans, reflecting a broader industry trend of equating AI with human capabilities.