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.
Tsinghua's HarnessPro cuts AI coding agent costs 26% while boosting accuracy without model retraining. Factory Droid 1.0 ...
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 ...
A new MIT CSAIL blog post is making the rounds in AI circles this week, and it describes something that sounds almost ...
Ariadne Loop - Local-first Loop Engineering workbench for writing verifiable AI coding-agent specs, verifier gates, rollback rules, and JSON reports. Cage - TypeScript contract harness that links ...
Astra identifies game policies from raw weight matrices and fixes reasoning loops in Qwen3.5-0.8B by editing 0.016% of parameters, lifting MATH-500 scores from 7.9% to 48.1% without training loops.
Foundry in the real world: Turning LLM hunches into verified findings – Security Loop Series, Part 2
See how Foundry turns LLM hunches into verified findings on Langflow, MLflow, and PyGoat, using an evidence gate and an honest look at token cost.
Article by Shenchao TechFlowNo training required, no data needed, no gradient descent.Researchers at MIT fed the weights of ...
Key Takeaways AI is moving beyond chat interfaces. Developers now build systems that use tools, retain context and take actions […] ...
MIT CSAIL finds GPT-6 Astra can read and write neural weights directly, fixing Qwen3.5-0.8B math from 7.9% to 48.1% with a 0.016% edit and no gradients.
Some results have been hidden because they may be inaccessible to you
Show inaccessible results