Models intelligence.

In an OpenAI case study, Legora tested Astra on a financial review involving 41 documents. It detected four deliberately inserted errors and completed the review in minutes.

OpenAI’s Playco case study describes three game prototypes built from a shared starting point. Playco reported half as many manual corrections as with the earlier model.

OpenAI introduces Astra with a focus on coding, computer operation, security and scientific work.

MIT News profiles three researchers whose work with the MIT–IBM Computing Research Lab led to roles at IBM. Their projects span reinforcement-learning agents, trustworthy AI and quantum computing, showing how academic collaborations can carry research into practical industry systems.

MIT and Motional researchers developed CW-Net, a system that explains an autonomous vehicle’s decisions through understandable concepts. In private-track tests and simulations, the explanations helped people anticipate vehicle behaviour and recognize mistakes. The findings concern these studies, rather than a demonstrated improvement in public-road safety.

OpenAI profiles ATV Big Air Tour’s use of ChatGPT for merchandise and marketing tasks, including building an inventory site from product photographs.

Google brings together its AI announcements from August in a monthly roundup.

OpenAI examines how Basis, Clay and Exa Labs apply agents to customer onboarding, account operations and software integrations.

OpenAI announces healthcare data connections for ChatGPT, bringing patient records and research into clinicians’ workflows.

OpenAI profiles Polimill’s work helping Japanese local authorities find administrative information using GPT models, with Codex supporting development.

OpenAI reports an annualized advertising revenue pace of $1 billion for ChatGPT and announces expansion into more markets.