Hypothesis: DeepSeek seeds on GPT data, and internal RL drives its more-extreme-than-GPT personality
teortaxesTex · x · 2026-10-29
teortaxesTex offers a hypothesis based on Artificial Analysis personality comparisons (DeepSeek V4.1 Flash scores far lower than GPT-6 Astra on Warmth, but is MORE extreme on Conviction traits like Confident, Constructive Pushback and Opinionated): DeepSeek may use GPT data as a seed, with most personality change carried by internal RL that the team doesn't yet know how to mitigate. He suggests this could explain why open models often feel more "Claude/GPT-slop" than actual Claude and GPT — though DS notably doesn't aggregate-match GPTs. Speculative, but consistent with the bench screenshots.
More from Models
- Insider teases Google still has unreleased SOTA-tier models in the chamber — iruletheworldmo · 2026-10-29
- AA-PersonalityBench: Grok 4.5 tops Answer-First at 100, Grok 4.6/4.7 rank #2/#3 on Directness — XFreeze · 2026-10-29
- GPT-6 Sol skips usage limits while other models burn them, unifying Chat and Work gets messy — Angaisb_ · 2026-10-29
- GPT-6.1 Terra scores 28.1% on agentic WeirdML v3, matching Claude Fable 5.1 at lower cost — scaling01 · 2026-10-29
- Opus 5.5 reportedly notably stronger than Astra at writing custom kernels — DimitrisPapail · 2026-10-29
- ChatGPT unifies Chat and Work — users wonder how usage limits will change — Angaisb_ · 2026-10-29