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.

MIT CSAIL's "weight-space programming" has a frontier LLM read and edit neural network weights directly, with no gradient descent. Surgical edits ended Qwen's reasoning loops, but the results come ...

Qwen3.5-0.8B, a small open-weight model from Alibaba, also saw GSM8K accuracy rise from 13.8% to 67.1% and IFEval from 21.0% to 62.3%. The researchers call the result early "sparks" of a capability, ...

Software engineering in all common languages and stacks, from small scripts to systems programming, including debugging, architecture, and review. Security research and engineering, including both ...

Can an LLM write a neural network that IS a benefits formula, exactly? Weight-space programming experiments on SNAP, EITC, SSI, and Medicaid arithmetic. - ctrimm/weight-space-programming ...

Terra, Claude Haiku 5.5, Gemini 4 Carbon Muse Wonder, Nano Banana 2.1, Genie 4. You open an announcement for a new model, you don't understand the terms used in the explanation and look up their ...