Get Started
Home
Topics
Search
Library
Research questionHow can language models recognize unknown entities and refuse to invent relational facts?Language models may generate plausible answers for entities absent from their knowledge instead of acknowledging that the information is unavailable. This makes it difficult to distinguish learned facts from fabricated relational completions.
AI
Alignment & Safety
Evaluation & Benchmarks
Natural Language Processing
Latest papersRecent research connected to this question, newest first.Relational Linearity is a Predictor of HallucinationsThe study examines synthetic entities designed to be unknown to the model across 15 relations, using the SynthHal benchmark and four instruction-tuned models. Its evidence is correlational: relational linearity predicts hallucination versus refusal, but does not establish the hypothesized causal mechanism.research paper · Sep 3, 2026
Related questions
How can we construct accurate knowledge bases from an LLM’s parametric knowledge without retrieval, fine-tuning, or exceeding 32B parameters?How can language models assigned protective roles avoid claiming real-world actions they cannot perform?How can language models perform complex logical reasoning without accumulating token-level errors?Can language models infer others’ mental states as social interactions evolve under unreliable information?