OpenAI is restructuring its safety leadership, with Johannes Heidecke departing and safety teams now reporting directly to new VP Mia Glaese, as the company pushes to embed safety work earlier and deeper into its frontier AI development process.
Google Cloud's new Discovery Bench framework exposes alarming flaws in AI benchmarking, revealing that a single vague word can crash an agent's performance score from perfect to zero, while also uncovering broken ground truth and mislabeled data in widely trusted benchmarks — raising urgent questions about whether the AI industry …
Chinese AI startup MiniMax Group is raising up to $2 billion through share sales and convertible bonds in Hong Kong, as the company sells 35.6 million shares at a 9.9% discount, sending its stock down nearly 9.8%, while simultaneously offering HK$6.5 billion in zero-coupon convertible bonds due 2027.
A new open-source AI framework called ACRouter launches on GitHub, intelligently routing coding tasks across multiple AI models to optimize both performance and cost, complete with a benchmark dataset, Hugging Face adapters, and plug-and-play integrations for popular developer tools.
Meta is set to begin production of its custom AI chips in September, developed with Broadcom and manufactured by TSMC, as the tech giant moves to reduce its dependence on Nvidia GPUs while projecting up to $145 billion in capital expenditures this year.
OpenAI retracts its support for SWE-Bench Pro after an internal audit uncovers that nearly 30% of the coding benchmark's 731 tasks are broken due to flawed tests and misleading prompts, prompting a call for new, developer-built benchmarks.
AI chip startup SambaNova surges to an $11 billion valuation after raising $1 billion led by General Atlantic, with JPMorgan Chase deploying its systems and a U.S. IPO targeted for 2027 as it challenges Nvidia's dominance in the AI chip market.
Chinese AI models like DeepSeek are surging past 30% U.S. market share as businesses abandon OpenAI and Anthropic for alternatives that are up to 90% cheaper and rapidly closing the performance gap with American frontier models.
DeepSeek is racing to develop its own AI chips after spending a year recruiting engineers and meeting with hardware partners, aiming to break free from dependence on both Nvidia and Huawei as Chinese and American AI giants alike pursue custom silicon for greater control over their technology.
A new study reveals that using JSON mode for AI agents causes smaller models to fail up to 60% of the time and costs up to 11x more than traditional CLI arguments due to shell escaping failures, with developers urged to avoid rewriting CLIs for agents until running measurable experiments.