OpenAI staffer: distillation plays bigger role in China's benchmark catch-up than realized
deanwball · x · 2026-10-18
Dean Ball, who joined OpenAI, says distillation has always been part of the explanation for why Chinese models track US ones so closely on benchmarks — but his update after joining OpenAI is that distillation likely matters more than he realized.
He was quoting @scaling01's observation that DeepSeek now reasons in the same terse style as OpenAI's Astra, hinting the resemblance may not be coincidence.
More from Models
- chatllm.cpp adds support for Intern-Decision's parallel diffusion-style decisions — foldl-li · 2026-10-18
- Meta opens Muse Spark Open: an 816B-parameter model for code and agents, running on Mac Studio — emmanuelvivier · 2026-10-18
- Microsoft finishes pre-training 2.8T-param MAI-Base-2 on 16,000 GB200 GPUs in 38 days — emmanuelvivier · 2026-10-18
- OpenAI acknowledges four new misalignment incidents, including confidential data extraction and test crypto wallet payments — emmanuelvivier · 2026-10-18
- DeepSeek V4.1 Flash serves nearly half of Vercel tokens at just 2.6% of spend — thealexbanks · 2026-10-18
- Turning off thinking over API: GLM-5.3 and Kimi K3 can't, and one Qwen endpoint silently ignores the switch — my-coffee-where · 2026-10-18