{"slug": "an-llm-edited-a-small-model-s-weights-directly-and-its-math-score-rose-6x", "title": "An LLM edited a small model's weights directly and its math score rose 6x", "summary": "MIT CSAIL researchers Rikiya Takehi, Jayden Teoh, Alex Zhang and Omar Khattab reported that a frontier LLM, GPT-6 Astra, directly edited 0.016% of Qwen3.5-0.8B's weights without any training run or gradients, lifting MATH-500 from 7.9% to 48.1% and GSM8K from 13.8% to 67.1%. The authors attribute most of the gain to unblocking: 93.0% of the base model's MATH-500 answers hit the 32,768-token limit without finishing, versus 33.1% after the edit, which removed a looping direction written by the MLP blocks of layers 15 and 19. The edited model and logs are public, though the approach showed no gain on Gemma 3 270M and the work is a blog post rather than a peer-reviewed paper.", "body_md": "**TL;DR**\n\n- GPT-6 Astra, a frontier LLM, edited 0.016% of Qwen3.5-0.8B's weights directly, with no training run. MATH-500 went from 7.9% to 48.1%.\n- The authors read most of that gain as unblocking: the base model kept running into its token limit instead of finishing. Astra also read, repaired and wrote small game-playing networks.\n- Limits: no gain on Gemma 3 270M, one frontier LLM, models up to 0.8B, a blog post and not a peer-reviewed paper, and a prompt that hinted at the loop. The edited model and the logs are public.\n\nOne edit to 0.016% of a small language model's weights lifted its math score from 7.9% to 48.1%. No training run, no gradients: GPT-6 Astra, a frontier LLM, read the weights directly and wrote the fix.\n\nRead the 6x with care. 93.0% of the base model's MATH-500 answers ran to the 32,768-token limit without finishing, and the authors read the gain as mostly unblocking: the model now finishes answers it could already work out. The assignment prompt also pointed Astra at that loop, and people have argued about it since Nov 2.\n\nMIT CSAIL researchers call this weight-space programming: a general-purpose LLM reads and writes a neural network's weights with no optimizer and no gradient descent, guided by a plain-language spec. Training has always been an indirect way to write weights, and their first results suggest a frontier model can write them directly. The post is \"Sparks of Weight-Space Programming\" by [Rikiya Takehi, Jayden Teoh, Alex Zhang and Omar Khattab](https://rikiyat.github.io/blog/weight-space-programming/).\n\n**The edited model is on Hugging Face as `RikiyaT/Qwen3.5-0.8B-weight-space-programming`, and only two tensors differ from base.**\n\nHugging Face's Ahsen Khaliq rebuilt it from the logs within a day and shipped a live demo. HF's independent run of the full MATH-500 set scored 51.0%.\n\n**Try this:** run the same math questions through base Qwen3.5-0.8B and the edited model with the same generation cap, and count how many answers hit the cap. The table below shows the authors' rates at a 32,768-token limit. Then clone the [logs repo](https://github.com/RikiyaT/lm_understand-logs) and open `exp8_math/runs/p50c/turns/t001/prompt.txt`. It is the exact assignment Astra received. Read it before any commentary, including this post.\n\nReplaying Astra's own session is the expensive part, though the logs hold everything sent to and from Astra, all code and all evaluations. Coverage of the logs (dailyaibrief.com, clauding.de) reports that the successful 50-pass run took 158 conversational turns and 80 Python runs over 20.3 hours. No fine-tuning cost is given to compare that with.\n\nQwen's own model card warns Qwen3.5-0.8B is prone to \"loops compared to other Qwen3.5 models, which may prevent it from terminating generation properly.\" Of the base model's first 96 practice answers, 85 ran to the 32,768-token limit.\n\nAstra worked in a Python notebook on practice prompts from the GSM8K and MATH training sets. A forward pass is one run of the model, and the budget capped how many runs Astra got: 10, 25 or 50. It traced the looping to one direction, a single pattern written by the MLP blocks of layers 15 and 19. It then subtracted that direction from 64 output columns in each block, so those layers no longer write it. That is 131,072 weights, 0.016% of the model, and the rest stays as trained.\n\n| Metric | Base Qwen3.5-0.8B | Edited (50 passes) | \n|---|---|---|\n| MATH-500 | 7.9% | 48.1% | \n| GSM8K | 13.8% | 67.1% | \n| IFEval | 21.0% | 62.3% | \n| MATH-500 answers hitting the 32,768-token limit | 93.0% | 33.1% | \n\n**It needed the full 50-pass budget: at 10 and 25 passes, MATH-500 reached only 12.3% and 14.1%.** Even at 50 passes, 33.1% of answers still hit the limit. IFEval, which tests instruction following, rose too, which fits the unblocking reading.\n\nA skeptic can call this a bug fix for a documented loop. None of the coverage I read tests whether a decoding-side fix, such as a repetition penalty, would match it.\n\nResearcher Lucas Beyer read the assignment prompt, as Turing Post reported on Nov 2. It told Astra that if Qwen \"repeats itself and never finishes, finding and fixing what drives that repetition counts.\" Coauthor Jayden Teoh replied that the team gave the short assignment via Claude Code.\n\nThere is a fair counterpoint. Qwen's model card had already flagged the loop, so pointing at it was reasonable.\n\n**The third-party re-run confirms the edit works. How much credit Astra deserves for finding the loop, given the hint, stays open.** A cleaner test would rerun the assignment with that sentence removed. None of the coverage I read reports one, and the prompt is public.\n\nGemma 3 270M rarely loops: 8.5% of its answers hit the token limit. MATH-500 started at 6.8% and four edits scored 7.0%, 7.6%, 7.7% and 7.0%. None was established, and Astra said so itself. Its errors are ordinary arithmetic and algebra mistakes, with no single component to switch off. The authors put it this way: \"Editing weights could remove an obstacle that hid an ability, but it could not add an ability the model lacked.\"\n\nIn a separate test on two fine-tuned Gemma copies, Astra failed from weights alone in all 4 attempts, so some hidden behavior shows only in activations.\n\nOne frontier LLM was tested, on models up to 0.8B parameters. This is a blog post with logs, not a peer-reviewed paper, and the authors themselves call it \"sparks\" and \"very early.\" They are building a benchmark and harness to make this \"a rigorous, reproducible engineering discipline.\"\n\n*Sources: numbers come from the authors' post and logs, plus the Hugging Face model card for the 51.0% re-run.*\n\nMy read: the practical use today is diagnosing and surgically repairing small models. If one of yours underperforms, check how often its answers hit the generation cap before you blame its capability. In Qwen's case, 85 of the first 96 practice answers hit it. Then ask whether a decoding-side fix would do the same job; none of the coverage I read reports that comparison.\n\nWatch for a replication above 0.8B parameters, and for a second model whose fault is one removable direction, which Gemma's was not.\n\nTraining is an indirect way to write weights, and a frontier LLM has now written some directly and beaten trained networks on small tests.\n\nDan writes the AI explainers on super-intelligence.rocks, where this piece first appeared.", "url": "https://wpnews.pro/news/an-llm-edited-a-small-model-s-weights-directly-and-its-math-score-rose-6x", "canonical_source": "https://dev.to/dan_lumatic/an-llm-edited-a-small-models-weights-directly-and-its-math-score-rose-6x-2d0l", "published_at": "2026-11-05 01:07:52+00:00", "updated_at": "2026-11-05 01:22:56.595353+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "machine-learning", "ai-research", "ai-tools"], "entities": ["MIT CSAIL", "GPT-6 Astra", "Qwen3.5-0.8B", "Rikiya Takehi", "Jayden Teoh", "Alex Zhang", "Omar Khattab", "Hugging Face"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/an-llm-edited-a-small-model-s-weights-directly-and-its-math-score-rose-6x", "markdown": "https://wpnews.pro/news/an-llm-edited-a-small-model-s-weights-directly-and-its-math-score-rose-6x.md", "text": "https://wpnews.pro/news/an-llm-edited-a-small-model-s-weights-directly-and-its-math-score-rose-6x.txt", "jsonld": "https://wpnews.pro/news/an-llm-edited-a-small-model-s-weights-directly-and-its-math-score-rose-6x.jsonld"}}