'Like gene editing for neural networks': MIT researchers used GPT-6 Astra to edit Qwen's open weights directly, boosting math performance

In the strongest example of "weight space programming", as the technique is called, Astra’s intervention raised Qwen3.5-0.8B’s MATH-500 score from 7.9% to 48.1% while touching approximately 0.016% of ...
GitHub

F1 Race Pace Forecasting

The frozen RF/B1 comparison uses saved predictions; analysis never refits or predicts again. Every group table includes lap/race counts, equal-race mean MAE, pooled lap MAE, gains and race wins/losses ...
eu.36kr

MIT allowed GPT-6 Astra to directly modify Qwen's weights, and its math scores have soared.

MIT has unveiled an innovative, groundbreaking paradigm dubbed "Weight Space Programming", representing a transformative new ...
kucoin

MIT researchers use GPT-6 Astra to enhance Qwen's mathematical ability by 6x

Article by Shenchao TechFlowNo training required, no data needed, no gradient descent.Researchers at MIT fed the weights of ...
en.sedaily

Google Legend Jeff Dean Bets on AI That Runs Its Own Experiments

Jeff Dean left Google after 27 years to found Discovery Loop, a startup automating research loops to run thousands of ...
X

Introduction to Reward Hacking in inference optimization

In inference optimization, reward hacking happens when an agent improves its score by exploiting the evaluation without delivering the improvement we actually wanted. This can happen at the kernel ...
kingy.ai

Weight-Space Programming: How GPT-6 Astra Raised Qwen’s MATH-500 Score Sixfold

GPT-6 Astra edited Qwen’s weights and raised MATH-500 from 7.9% to 48.1% in author-run tests. How weight-space programming works—and its limits.
51CTO

不用换大模型！谷歌 Harness² 双层递归，让 Agent 实现自我迭代升级

本文讲清 Harness² 解决什么问题、机制怎么跑通、实测数据、社区落地与未来边界。 做 Agent 的同学大概都有过这种感受：同一个大模型，换一套提示词、工具编排和记忆策略，干活稳定性就能差 ...