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.
retained text
No document body was retained for this record — only its metadata was stored.
document fetched 2026-11-03 00:30 · canonical https://reddit.com/r/singularity/comments/1xm89o3/mit_researchers_say_gpt6_astra_improved_a_small