PrefEval: preference-following accuracy falls below 10% at just 10 turns for most LLMs
math_rachel · x · 2026-10-19
From Rachel Thomas's thread, quoting the PrefEval benchmark (arXiv:2502.09597): in zero-shot settings, preference-following accuracy falls below 10% at merely 10 turns (3k tokens) across most evaluated models. Even advanced prompting and retrieval methods keep deteriorating in long-context conversations. The benchmark has 3,000 curated preference-query pairs across 20 topics, tested 10 open and proprietary LLMs up to 100k tokens, and shows fine-tuning on PrefEval significantly improves performance.
More from Research
- Dream2Discern: Self-Generated Hard Examples Predict Multi-Agent Failures With 90%+ Accuracy — RutgersU · 2026-10-19
- BridgeWAM: latent bridge queries decouple video and action experts, 2.47x faster robot inference — Fudan-University · 2026-10-19
- NVIDIA's FC-SWE: Failure-Conditioned RL Boosts Multi-Attempt Resolution for SWE Agents — nvidia · 2026-10-19
- AVC-Reward: an event-centric framework for evaluating audio-video generation consistency — ZhejiangUniversity · 2026-10-19
- PremiseCheck Fact-Checks Peer Reviews at GPT-5.5-Level Accuracy for Half the Cost — Dingjie Song · 2026-10-19
- SciΠvot benchmark: frontier models prefer flawed research plans and lag human scientific judgment — Youran Sun · 2026-10-19