Trending storyDeveloping
Astra weight edits lift small model's math scores
1 report1 sourceupdated 2 hours ago
What happened
AI digest
On November 3, Hacker News featured a report on Weight-Space Programming research. The report says GPT-6 Astra improved Qwen3.5-0.8B's math performance by directly editing its weights, changing 0.016% of them. It reports Qwen3.5-0.8B's MATH-500 score rose from 7.9% to 48.1%. The coverage only describes the research report and its published results, with no follow-up verification or further progress.
Written by AI from the coverage · updated 1 hour ago
Coverage
Follow the reports to see the story from different sides.
Nov 3
- Hacker News front pageAstra can read and write weights to improve Qwen
Weight-Space Programming 研究报告称，GPT-6 Astra 直接编辑 Qwen3.5-0.8B 的 0.016% 权重，将 MATH-500 从 7.9% 提升至 48.1%。
Heat over time
Not enough continuous observations to draw a trend yet.