AI Video Models Fail Their Own Physics Exam
A new benchmark tested eight AI video models on physics accuracy. The best scored 57.76 out of 100, exposing a core reliability gap.
AI news and analysis covering the labs, models and infrastructure behind the industry: what OpenAI, Anthropic, Google DeepMind, Meta and Mistral ship, and what each release changes for the people building on top of it.
A new benchmark tested eight AI video models on physics accuracy. The best scored 57.76 out of 100, exposing a core reliability gap.
Claude Models: Anthropic expanded its Cyber Verification Program, giving verified security professionals broader access to Claude’s most capable AI models for defense.
Apple is working with LG on seven smart home devices, including a doorbell and locks, to challenge Amazon and Google’s home hardware dominance.
Southwest Airlines launched the first US airline plugin inside ChatGPT on Oct. 6, working with OpenAI and AWS on the build.
More than 33 million Americans who do not own crypto are likely to buy it in 2026. Key Takeaways Within that 59%, 7% expect to buy soon, 11% are waiting for the right moment, 17% are curious but unsure where to start and 24% call themselves skeptical but open-minded. The remaining 41% actively avoid crypto….
Nearly Double: A new finding shows two filler words can lift a base AI model’s math accuracy from 42% to 78%, exposing a training data quirk.
Le Chonk: Mistral released Mistral Large 4, a 1 trillion parameter open-weight model with 49 billion active parameters, rolling out via API Oct. 6.
Hyperscalers will spend $775-800B on AI infrastructure in 2026. Fathom examines why grid capacity, not chips, is now the real constraint.
A new ETF called MN has filed to give retail investors indirect access to private AI firms OpenAI and Anthropic, among six target companies.
Rep. Ro Khanna plans a Data Center Bill of Rights to address AI data center energy demands, building on the federal AI Bill of Rights framework.