MIT CSAIL reports modification of neural weights in GPT-6 Astra
The story so far
MIT CSAIL claims GPT-6 Astra can explain and modify neural network weights without training loops. In a Qwen test, the authors stated that a small weight change significantly improved math and instruction following scores, but this finding is preliminary. The project, involving Andrej Karpathy, aims to provide a reading experience different from traditional browsing, released on the Hacker Times platform.
AI-written from the reports · Updated 15 days ago
Coverage timeline
Follow the coverage to see every side of the story.
- Hacker News · 300+Show HN: Hacker Times, a New York Times-inspired version of Hacker News
Hacker Times is laid out in a newspaper style to present Hacker News stories, linking to the original sources and updating every hour. The project, involving Andrej Karpathy, aims to provide a reading experience different from the traditional browsing experience.