/Tech1h ago

Researcher Surprised by RLVR Effectiveness in Complex Mathematics

2000650

Original post

Thomas G. Dietterich@tdietterich#62inTech

@andrewgwils I'm surprised at the effectiveness of RLVR for mathematics (if that is what they are using). I would love to see a detailed analysis of what knowledge is required at each stage of these long proofs. My assumption has been that LLMs can learn to instantiate general rules ...

Andrew Gordon Wilson@andrewgwils

Keep in mind that the current skeptics would have overwhelmingly said three years ago that many of the capabilities we are seeing now (e.g. solving important open problems in mathematics) would not have been achieved through these approaches.

8:28 PM · Jun 17, 2026 · 348 Views

Sentiment

Sentiment building, check back later.

Cluster Engagement

Digg Deeper

No Digg Deeper questions have been answered for this story yet.

Posts from X

Most Activity

VIEWS246REPLIES1

Thomas G. Dietterich@tdietterich

@andrewgwils if they are given enough examples. And that they can memorize unusual special cases from very few (maybe just 1) examples. Are those two kinds of knowledge enough? Or is there an intermediate level that is noisy but can be overcome with enough test time compute...

Thomas G. Dietterich@tdietterich

1h24600

Thomas G. Dietterich@tdietterich

@andrewgwils Can anyone recommend pointers to relevant work?

Thomas G. Dietterich@tdietterich

1h5600