5:00 PM
Today in brief.
Odyssey-3 tops the physics tests for world models
Odyssey says its new world model scored highest on World Models' Last Exam in Physics and remains state of the art on Physics-IQ Verified.
🔬 Science
Tutor Intelligence rents factory robots by the hour and teaches them with remote humans
a16z led Tutor Intelligence's Series B: robots deploy before they are fully autonomous, and a remote human 'tutor' takes over when one gets stuck, which also creates training data.
🔧 Tech
Shoppers sent by AI convert 60% better
a16z's chart of the week says shoppers referred by AI show up ready to buy: higher conversion, higher revenue per visit and fewer bounces than other online shoppers.
2 sources 3:43 PM
💼 Business
Karpathy open-sourced the data behind HackerTube
Andrej Karpathy had Claude build a YouTube-style site of 6,454 videos recommended in Hacker News comments in about 2 hours, then released the dataset so anyone can build their own version.
2 sources 3:38 AM
🔧 Tech
Try today: Runway inside Cursor, a free 12B thinking model, Kimi for 3D
A short list of things you can open today.
+15 sources 🤖 🤖 AIGPT-6 Astra raised a small Qwen model's math score by editing its weights
A paper reports GPT-6 Astra lifted a 0.8B Qwen model from 7.9% to 48.1% on MATH-500 by editing about 0.016% of its weights directly, and a sharp skeptic says a human researcher would have done the same fix.
🏛️ Policy
Anthropic's filing says it may imminently hit its own self-improvement threshold
An amended S-1 says internal estimates suggest Anthropic may soon meet its Responsible Scaling Policy definition of dramatically accelerated progress attributable to AI, and that it might not notice until after it happens.
💼 Business
Anthropic's Q3: $17B revenue, growth slowing, internal AI use above payroll
The amended S-1 gives Q3 numbers: revenue up 47% quarter on quarter, down from 144% in Q2, and internal Claude use valued at retail prices now exceeds employee pay.
No stories in this category for this edition.