AI News Brief

AI News Brief: BenchMIRT is a new method for breaking down what LLM benchmarks actually measure, auditing them at the individual question level. Traditional benchmarks usually claim to measure a single capability (like safety, general reasoning, or instruction following), but individual questions often involve multiple capabilities at once. For example, BBQ, a benchmark that tests whether models rely on social stereotypes, has a question about a grandparent and grandchild booking an Uber that looks like it’s testing age bias, but actually also requires the model to correctly track character roles and reason from evidence rather than assumptions.

AI News Brief: This news comes from the Google DeepMind blog, not internal data, translated directly: Google has launched a new speech-to-text model, Gemini 3.5 Transcribe, built for accurate and intelligent speech recognition, already powering the Gemini App, Android Rambler, and the macOS Gemini App…

AI News Brief: OpenAI’s latest report explores how students and educators are using ChatGPT to extend learning beyond the classroom, turning it into an ongoing experience that isn’t tied to a fixed time or place. See the original article for full details.

AI news brief: AI hedge fund Situational Awareness is being investigated by the SEC, with banking records subpoenaed. Situational Awareness, an AI-focused hedge fund led by former OpenAI employee Leopold Aschenbrenner in his mid-20s, recently fell into serious crisis. The fund made big bets across multiple AI-related investments and once saw stunning growth, but a sharp AI stock selloff in late July caused the company’s…

AI news brief: A GitHub blog post explores how a ‘canvas’ interface can be used in agentic workflows to address the limits of pure chat interactions. The author points out that while chat interfaces are great for expressing intent, multi-step agent tasks tend to get buried in scrolling conversation history, making them hard to track and control. The post shares how the author actually integrated a canvas into their own agentic workflow…

AI news brief: Stripe has reportedly finalized a deal to acquire OpenRouter for over $7 billion according to Bloomberg’s latest report, after the Wall Street Journal first revealed the two companies were in talks. OpenRouter helps customers route across multiple AI models based on task needs and budget, avoiding lock-in to a single provider.

AI news brief: Unitree, a leading Chinese humanoid robot maker, triggered a wave of retail over-subscription ahead of its listing on Shanghai’s STAR Market, drawing 9.78 million online applications with a final allotment rate below 0.0181%, the lowest on record for the STAR Market. The frenzy highlights how China’s tech narrative is shifting from AI models themselves toward the physical systems that put AI to work, with humanoid robots becoming one of the biggest investment themes even though commercialization is still early — a shift the US already treats as part of its strategic competition with China, having banned imports of Chinese-made humanoid robots over national security and cybersecurity concerns.

AI News Brief: Anthropic has reportedly struck a roughly $9 billion, 20-year deal with bitcoin miner Riot Platforms, under which Riot’s Rockdale, Texas campus will supply 191 megawatts (MW) of compute capacity for Anthropic’s AI operations. The news first surfaced in a Riot press release on Monday, which did not name the client, but Bloomberg cited sources confirming the partner is Anthropic; Cointelegraph has reached out to both companies for comment but has not yet received a response.

AI News Brief: Greg Abel, successor to tech-investing giant Warren Buffett, will travel to Japan next month to meet with leaders of the country’s major trading houses for the first time in his role as CEO of Berkshire Hathaway. The trip is seen as a major milestone in Berkshire’s longstanding relationship with Japan’s five biggest trading houses (Itochu, Marubeni, Mitsubishi Corporation, Mitsui & Co., and Sumitomo Corporation). Once mere trade intermediaries…

AI News Brief: OpenAI CEO Sam Altman shared what he called a ‘cool idea’ on social media last Friday — parents can link their family calendar to the company’s new product ChatGPT Work, tell it about their kids’ interests, and have it auto-generate a podcast every morning during the school run covering one kid’s soccer game that afternoon, another kid’s upcoming birthday, and some news.