AI research meets outside scrutiny
OpenAI announced three AI-generated research results spanning mathematics, computer science, and physics. The company says its models produced a proof of the local smoothing conjecture in every dimension, refuted the Exponential Time Hypothesis, and developed an existence proof for certain four-dimensional quantum field theories.
OpenAI describes the manuscripts as raw research data rather than settled scientific understanding. External researchers will examine the computer-science and physics work; the natural-language proof and Lean formalization for the complexity result are public, while collaborators have deferred release of the physics artifacts.
Share recurring ChatGPT work with a team
Team tasks in ChatGPT let colleagues maintain recurring work together. A task can prepare a weekly project update, meeting brief, or digest from approved sources, then run in the cloud on a schedule or supported trigger.
The feature is available in Business and Enterprise workspaces. Tasks use saved instructions and approved workspace connections, while authorized teammates can review results and update instructions when the work changes.
Also
- Let Copilot approve clearly safe requests Assisted permissions are generally available in Copilot CLI, the Copilot app, and Visual Studio Code; requests not classified as clearly safe still require approval.
- Automate logistics without handing over the creative work Aaron Makelky explains how a Notion command creates social-calendar entries from a producer’s project document while he keeps the writing for himself.
- Map neurons in an AI literacy game Google and the Stanford Accelerator for Learning created Polar Peaks for students ages 11 to 14, with tasks involving data cleaning, model testing, and human oversight.
- Build an app one complete slice at a time Aishwarya Srinivasan’s Codex walkthrough uses a job application tracker to show planning, architecture review, persistence, server-side validation, and testing.
Pass it on