HOT RADAR
llmHEAT 12.5°

AI CURATED HOT STORY · 11/19/2026

Editing Just 0.016% of Weights Cures Model Reasoning Loops

1 reports archived1 independent sourcesupdated 11/19/2026, 23:32:47
Synthesis & Latest Updates
1 Sources Cross-Validated

The author built an interactive WebGPU walkthrough of "weight-space programming": without gradient descent, editing only 0.016% (131,072 of ~800M) of a small Qwen model's ffn_down output weights across layers 15 and 19 largely cures its self-verification loop. On identical prompts, runs hitting the 32k-token cap fell from 93.0% to 33.1%, while MATH-500 accuracy rose from 7.9% to 48.1%. Unlike broad model souping, the method targets a few specific failure cases and the neurons behind them, and is pitched as a fit for agentic propose-test-feedback loops.

LATEST/The author built an interactive WebGPU walkthrough of "weight-space programming": without gradient descent, editing only 0.016% (131,072 of ~800M) of a small Qwen model's ffn_down output weights across layers 15 and 19 largely cures its self-verification loop. On identical prompts, runs hitting the 32k-token cap fell from 93.0% to 33.1%, while MATH-500 accuracy rose from 7.9% to 48.1%. Unlike broad model souping, the method targets a few specific failure cases and the neurons behind them, and is pitched as a fit for agentic propose-test-feedback loops.

Hourly heat curve

2 fully observed hours
Fewer than 3 fully observed hours; trend not plotted yet.

Coverage timeline

Total 1 reports · Latest first