The AI knowledge 99.9% of people never discover
SimLife tracks 106 episodes of simulated daily life and finds frontier AI models guess habits by frequency, not reasoning.
Together AI's Marek and Ryabinin show score centering cancels drift in reinforcement learning of LLMs from 0.6B to 30B parameters.
OpenAI's latest economic research claims workers use AI beyond their traditional roles, but the full report and its data remain unpublished.
A community-maintained directory of Claude Skills has crossed 75,000 GitHub stars, but its production-ready claim is a self-report, not an audit.
Scaling exponents, long thought fixed by compute and data, can be shifted by architecture changes like model growth, affecting how labs train large
OpenAI's September 16 guide urges firms to measure AI value with ChatGPT Work and Codex analytics, with no customer evidence or pricing.
OpenAI is building Sponsored Agents and marketer tools with HubSpot and Shopify, extending its ChatGPT ads pilot into a broader ad platform.
An arXiv decision model separates AI piloting from committing, showing when rapid progress makes waiting the rational enterprise choice.
The open-source Python tool promises to generate finished short videos from a keyword, but every capability claim is self-reported by its maintainer.
