proomt

Search

Search posts, papers, and topics

All posts

Hugging Face Daily PapersNiket Patel, Ahmad Rammal, Amaury Hayat1 min readpaperadvanced

Learning to Discover Interesting Mathematics

Read on arxiv.org

Summary

This paper proposes a metric for "interestingness" in mathematics, defined as the ratio of proof length to statement length, which correlates with a theorem's utility. They trained a 27B model to predict proof difficulty, showing it can generate more novel and interesting theorems by optimizing for this metric.

  • Intrinsic interestingness is defined as the ratio of proof length to statement length.
  • This intrinsic metric strongly correlates with the extrinsic downstream utility of a theorem.
  • A 27B model was trained to predict proof difficulty, outperforming frontier general-purpose models.
  • Optimizing for this metric reduced overlap with Mathlib from 91.9% to 30.6%, generating more novel math.

This work is significant for AI researchers and mathematicians seeking to leverage LLMs for guided, novel mathematical discovery without relying on human-supplied targets.

8/10
llmmachine learningaimathematicsresearch

Related reading

  1. If math is more than proof, we need to better celebrate the rest of it

    The author argues that mathematics should reward “motivated explanations” – narrative, intuition‑driven expositions that clarify why a theorem is interesting and how it fits into broader context – on par with traditional proof‑oriented work. He defines the concept, contrasts it with proofs, cites examples (Princeton Companion, Thurston’s essays, Chow’s exposition paper), and suggests formalizing…

    Hacker News front pagewordpress.com12 minHN425284

  3. When2Think: Learning Difficulty-Aware Length Control for Efficient Hybrid Reasoning Models

    When2Think introduces a post‑training framework that lets a large reasoning model decide per‑instance how much reasoning depth to allocate, using difficulty‑aware reward shaping (IDAC) and verifier rewards. It cuts token usage by ~28% while boosting Pass@3 by 10% on AIME24 and reaches 40% Pass@3 on AIME25, outperforming compression and routing baselines.

    Hugging Face Daily Papersarxiv.org1 minpaper

  6. Can We Trust the Teacher? Decoupled Credit Direction-Magnitude for Self-Distillation

    Existing self-distillation methods couple credit direction and magnitude, making them vulnerable to teacher errors and preference variance. Decoupled Credit Self-Distillation (DCSD) addresses this by theoretically separating credit direction and magnitude into two reliable signals, using belief-margin probing and marginal information gain to calibrate teacher supervision.

    Hugging Face Daily Papersarxiv.org1 minpaper