Absolutely incredible: MIT researchers say GPT-6 Astra improved a small Qwen model’s math score from 7.9% to 48.1% by directly editing just 0.016% of its weights, without gradient-based training.
The result points toward AI systems that can diagnose and improve other models directly, a potential step toward self-improving AI.
BREAKING: The OpenAI employee who was arrested and charged with allegedly using a hidden recording device in one of the ChatGPT maker’s San Francisco offices had a one-way ticket to Saudi Arabia booked for October 12, per Bloomberg.