Ben Clavié

@bclavie

regressing linearly on a daily basis. wife guy who does retrieval. research @mixedbreadai, prev answerdotai - dms work randomly, responses might be very slow

following
followers

GPT-6 Astra reportedly boosts Qwen3.5-0.8B’s MATH-500 score by editing 0.016% of its weights

The reported MATH-500 score rose from 7.9% to 48.1% with the same sampling settings.

226 Sources, 6d ago, updated 6d ago

FreshStack website and leaderboard revamp announced with newer retrieval models

In an Oct. 22 post, beirmug listed Cohere Embed v5, Google's EmbeddingGemma2 and Perplexity late-interaction models.

8 Sources, 20d ago, updated 20d ago

The challenge of building AI benchmarks from human tasks

A post argues that human tasks are now too easy to use as AI benchmarks in many domains.

58 Sources, 17d ago, updated 17d ago

Featured in

GPT-6 Astra reportedly boosts Qwen3.5-0.8B’s MATH-500 score by editing 0.016% of its weights

FreshStack website and leaderboard revamp announced with newer retrieval models

The challenge of building AI benchmarks from human tasks

Featured in

GPT-6 Astra reportedly boosts Qwen3.5-0.8B’s MATH-500 score by editing 0.016% of its weights

FreshStack website and leaderboard revamp announced with newer retrieval models

The challenge of building AI benchmarks from human tasks