BREAKING
AI loop cracks 9 open CS problems
0
COLT open list
0
FOCS 2023
0
commutative algebra
How the AI math harness works
1
GPT-5.5 Pro proves
↓
2
Claude Opus 4.8 verifies
↓
3
Iterate loop
↓
4
Humans check
Harness engineering, not new models
This pipeline
natural language
●
Prover-verifier loop
●
Stress-tested on COLT and FOCS
●
No familiarity needed
Prior methods
formal
●
AlphaProof verification
●
FunSearch search
●
Domain-specific agents
Researchers shift their view
Focus turns to broader verification
AI NEWS BLITZ
A prover-verifier loop of two frontier models reportedly solved nine open theoretical problems.