/Tech4h ago

NMSLIB creator Leo Boytsov argues frontier math AI claims must demonstrate higher pass@k rates than IMO benchmarks

OpenAI and Google models solved Erdős problems but failed IMO.

1300217

#95

Original post

Leo Boytsov@srchvrs#560inTech

@yoavgo Unless you have a consistently higher pass@k rate on frontier math problems compared to IMO problems I wouldn't be making concessions.

(((ل()(ل() 'yoav))))👾@yoavgo

because the erdos problem solutions could rely on fuzzy retrieving combining niche knowledge from different fields of human expertise, while IMO problems require proper skill and technique

9:40 PM · Jun 15, 2026 · 131 Views

Sentiment

Users affirm the AI researcher's questioning of any edge on Erdos problems versus IMO tasks as a legitimate point.

Pos

100.0%

Neg

0.0%

1 comments with sentiment.

Cluster Engagement

Posts from X

Most Activity

(((ل()(ل() 'yoav))))👾@yoavgo

@srchvrs legit

Leo Boytsov@srchvrs

@yoavgo Unless you have a consistently higher pass@k rate on frontier math problems compared to IMO problems I wouldn't be making concessions.

4h9100