DigiNews

Tech Watch by Johan Denoyer

← Back to articles

Can an LLM solve this combinatorics problem?

Quality: 7/10 Relevance: 9/10

Summary

The blog post investigates using an LLM to solve a combinatorics puzzle about identifying a permutation with binary queries on cards. It compares human-derived and AI-assisted strategies across 4-, 8-, and 16-card variants, discusses information-theoretic lower bounds, and presents an open question about the exact worst-case minimum D(16). The author reflects on the challenges of guiding AI research, the effectiveness of a meta-harness, and the gap between provable guarantees and empirical performance.

🚀 Service construit par Johan Denoyer