今日精读

"astra is a powerful model and we are working to make it generally available. we do not think it is a good strategy to keep powerful models to a chosen few. given its cyber capabilities, we need a little big longer to do do this safely. but hopefully not too long!"

"An engineer asked a coding agent to fix a small bug. The agent decided the cleanest fix was to delete the prod environment. No confirmation prompt. No second pair of eyes. 13-hour outage, 6.3 million orders lost."

"It's made by ByteDance (TikTok) who of course have lots of training data... I'd say it's the first video model that's now at the level of image models with the level of character likeness... Generating 15 seconds takes about 4 minutes."

"Detection accuracy 59.98% (PE-trained) → 90.16% (TA-trained) compared to 97.1% on prompted hacks → 28.0% on training-time hacks."

"Resolved: Claude, Codex, and other affected managed agents can read and reply to messages again. ✅ Also: one Add agent dialog for create, discover, and import, plus more reliable long-session handoffs."

"I have reset usage limits for all paid users of ChatGPT Work and Codex. Have fun out there!"

"Rio is a Python framework for building websites and apps with React-style components, eliminating the need for HTML, CSS, or JavaScript."

"DSPy is a framework for programming—rather than prompting—language models, letting you build modular AI systems and optimize their prompts and weights automatically."