Research · 4 November 2026
ResearchMIT researchers use LLMs to edit neural network weights, boosting MATH-500 score
MIT researchers introduced Weight‑Space Programming, letting GPT‑6 Astra edit 0.016% of a model’s weights and raise its MATH‑500 score from 7.9% to 48.1%.
Benchmarks are the company's own; not independently verified.
Why it matters
Benchmark model‑editing workflows using Weight‑Space Programming; users will expect rapid capability upgrades without full retraining.
For product managers
For your product
What does this mean for your product?
Try it on your product More from this issueAll AI news from 4 Nov →
Get this every morning, written for your product.
Deltaday reads the whole AI news cycle and tells you what changed, whether it's real yet, and what to do about it.