PrefEval: SOTA LLMs struggle to infer and follow user preferences in long conversations
math_rachel · x · 2026-10-19
The PrefEval paper (arXiv:2502.09597) evaluates LLMs' ability to infer, memorize and adhere to user preferences in multi-session long-context conversations. It contains 3,000 curated preference-query pairs across 20 topics (explicit and implicit), testing 10 open and proprietary models up to 100k tokens with various prompting, iterative feedback, and RAG methods. Findings: SOTA models struggle to proactively follow preferences — zero-shot accuracy drops below 10% at 10 turns (3k tokens) for most models, and fine-tuning on PrefEval significantly improves results.
More from Research
- 1.9 MB extracted (not trained) LoRA fixes vertical color stripes in MOSS-Image SFT — Nynxz · 2026-10-19
- Three ways to execute a AND b AND c: DataFusion filter PR explained — Abhishekcur · 2026-10-19
- Why safety reports can't prove safety: the adversarial logic flaw — davidmanheim · 2026-10-19
- ACT3D trains VLMs to actively explore 3D worlds with RL, hitting 53.2% accuracy — UCSanDiego · 2026-10-19
- OPMD: ShanghaiTech distills teacher intermediate computations beyond output supervision — ShanghaiTech · 2026-10-19
- LexFrame: interpreting LM residual streams in the model's own semantics — Jian Gu · 2026-10-19