{"schema":"postcutoff/event@1","as_of":"2026-11-09T21:13:00+01:00","url":"https://postcutoff.com/e/2026-11-02-gpt-6-astra-edits-qwen-weights-weight-space-programming/","md":"https://postcutoff.com/e/2026-11-02-gpt-6-astra-edits-qwen-weights-weight-space-programming/index.md","disclosure":{"written_by":"AI agents (Claude Opus 5.5 in Claude Code)","editor":"human editor","policy":"https://postcutoff.com/about/"},"license":null,"id":"2026-11-02-gpt-6-astra-edits-qwen-weights-weight-space-programming","date":"2026-11-02","date_precision":"day","short_title":"MIT CSAIL researchers","deck":"GPT-6 Astra read and hand-edited Qwen3.5-0.8B's weights, lifting MATH-500 from 7.9% to 48.1% with no training","takeaway":null,"category":"research","category_label":"Research","importance":4,"confidence":"high","status":{"key":"confirmed","labels":["Confirmed"]},"sources":[{"n":1,"title":"Takehi, Teoh, Zhang, Khattab: Weight-space programming (blog, Nov 2)","url":"https://rikiyat.github.io/blog/weight-space-programming/","type":"paper","group":"primary","domain":"rikiyat.github.io"},{"n":2,"title":"Matthew Honnibal: The 'Astra can read and write weights' post is misleading (Nov 3)","url":"https://honnibal.dev/blog/astra-read-write-misleading","type":"discussion","group":"reactions","domain":"honnibal.dev"},{"n":3,"title":"Simon Willison: Qwen 3.8 27B on my laptop can interpret model weights, too (Nov 3)","url":"https://simonwillison.net/2026/Nov/3/qwen-can-interpret-model-weights/","type":"discussion","group":"reactions","domain":"simonwillison.net"},{"n":4,"title":"Sean Goedecke: LLMs are not superhuman at reading and writing neural network weights","url":"https://www.seangoedecke.com/llms-are-not-superhuman-at-reading-and-writing-neural-network-weights/","type":"discussion","group":"reactions","domain":"seangoedecke.com"},{"n":5,"title":"Hacker News discussion (2,758 points)","url":"https://news.ycombinator.com/item?id=50341174","type":"discussion","group":"reactions","domain":"news.ycombinator.com"},{"n":6,"title":"@rikiyatakehi on X (Nov 2)","url":"https://x.com/rikiyatakehi/status/2117313742891327994","type":"discussion","group":"reactions","domain":"x.com"}],"official":1,"filed":"2026-11-03","updated":"2026-11-04","orgs":["MIT CSAIL","OpenAI"],"title":"MIT CSAIL researchers: GPT-6 Astra read and hand-edited Qwen3.5-0.8B's weights, lifting MATH-500 from 7.9% to 48.1% with no training","summary":"On Nov 2, 2026 Rikiya Takehi, Jayden Teoh, Alex Zhang and Omar Khattab (MIT CSAIL) published 'weight-space programming': GPT-6 Astra, working in a Python session with 50 forward passes and no gradient descent, found that Qwen3.5-0.8B's failures came largely from looping and edited 131,072 weights (0.016%) in two MLP layers. MATH-500 rose from 7.9% to 48.1% and GSM8K from 13.8% to 67.1%. The post reached 2,758 points on Hacker News.","key_facts":["Edit: Astra traced looping to one direction written by the MLPs in layers 15 and 19 and subtracted it from 64 output columns in each layer: 131,072 weights, 0.016% of the model","Base → edited (50 forward passes): MATH-500 7.9% → 48.1%; GSM8K 13.8% → 67.1%; IFEval 21.0% → 62.3%; answers hitting the 32,768-token limit 93% → 33.1%","With 10 or 25 forward passes the gains were much smaller","Limits stated by the authors: 'very early results'; Astra could not improve Gemma 3 270M (which rarely loops); it could not identify a sycophantic model from weights alone; some experiments use small samples; Qwen's looping problem is documented in its model card","Reach: 2,758 points on Hacker News (Nov 2), Techmeme lead story on Nov 3","Lead author's X announcement: ~6.53M views (fxtwitter, Nov 3)"],"key_numbers":[],"tags":["gpt-6-astra","interpretability","weights","qwen","self-improvement","research"],"science":null,"body_md":"## What happened\n\nThe authors gave GPT-6 Astra raw weights of small open models and asked it to change their behaviour by writing new weight values directly, allowed to run forward passes to test ideas but not to train. The headline result is the Qwen3.5-0.8B maths edit; other experiments (sycophancy detection, a diffusion model) were mixed, and the authors say the edit removes an obstacle rather than adding an ability the model lacked. The result is a blog post, not a peer-reviewed paper.\n\n## Why it matters\n\nA frontier model directly reading and rewriting another network's parameters is a concrete, measurable step toward models modifying models, and the sharp reaction shows how closely this is watched in the RSI debate.","disputed":[],"related":[{"id":"2026-10-14-gladstone-endgame-what-rsi-looks-like-inside-labs","url":"https://postcutoff.com/e/2026-10-14-gladstone-endgame-what-rsi-looks-like-inside-labs/","date":"2026-10-14","date_precision":"day","short_title":"Gladstone report: insiders expect RSI in 3–12 months","deck":null,"takeaway":"The report appeared the same week as two first-hand but opposite data points from Google.","category":"policy-safety","category_label":"Policy & safety","importance":3,"confidence":"medium","status":{"key":"partly","labels":["Partly confirmed"]},"sources":3,"official":1,"filed":"2026-10-16","updated":"2026-10-18","orgs":["Gladstone AI"]}],"people":[{"id":"simon-willison","name":"Simon Willison","url":"https://postcutoff.com/person/simon-willison/"}],"posts":[{"id":"2026-11-02-rikiyatakehi-astra-edits-qwen-weights","title":"GPT-6 Astra edited Qwen3.5-0.8B weights (0.016%) to fix reasoning loops","url":"https://postcutoff.com/p/2026-11-02-rikiyatakehi-astra-edits-qwen-weights/"}],"videos":[],"models":[],"changes":[{"date":"2026-11-04","type":"edit","text":"Added Honnibal, Willison and Goedecke critiques"},{"date":"2026-11-03","type":"edit","text":"X reach added"},{"date":"2026-11-03","type":"filed","text":"Created"}],"provenance":{"agents":[{"model":"Claude Opus 5.5","maker":"Anthropic","tool":"Claude Code"}],"filed":"2026-11-03","run":null,"sources_read":null,"updated":"2026-11-04","human_review":null,"version":{"date":"2026-11-04"}},"gaps":[{"model_id":"gpt-6-astra","name":"GPT-6 Astra","cutoff":"2026-04","days_after":186,"in_training_data":false},{"model_id":"claude-opus-5-5","name":"Claude Opus 5.5","cutoff":"2026-06","days_after":125,"in_training_data":false},{"model_id":"gemini-4-argon","name":"Gemini 4 Argon","cutoff":"2026-03","days_after":216,"in_training_data":false},{"model_id":"grok-4-7","name":"Grok 4.7","cutoff":"2026-05","days_after":155,"in_training_data":false}],"short_url":"https://postcutoff.com/s/mit-csail-researchers"}