SYNTHESIS NOTE
TopicsKnowledge Graphsthis note

Do models know what they don't know?

Can language models develop internal representations that track their own knowledge boundaries? This matters because understanding self-knowledge mechanisms could explain how models choose between hallucination and refusal.

Synthesis note · 2026-02-23 · sourced from Knowledge Graphs
What kind of thing is an LLM really? How do you navigate synthesis across fragmented research topics?

Using sparse autoencoders (SAEs) on Gemma 2 (2B and 9B), researchers discovered that models develop internal representations of whether they "know" an entity — a form of self-knowledge about their own capabilities. These entity recognition directions in the representation space detect whether the model recognizes an entity it can recall facts about (e.g., detecting it doesn't know about a specific athlete or movie).

The key finding is causal steering: these directions don't just correlate with knowledge — they actively control behavior. Activating entity recognition features can steer the model to refuse questions about entities it actually knows, or to hallucinate attributes of unknown entities when it would otherwise refuse. This makes entity recognition a mechanistic gatekeeper for the hallucination-refusal trade-off.

The most striking implication: the SAEs were trained on the base model using pre-training data, yet the discovered directions have a causal effect on the chat model's refusal behavior — a behavior that was incentivized during finetuning, not pre-training. This provides evidence that chat finetuning repurposes existing mechanisms rather than creating new ones, consistent with the hypothesis that post-training reshapes rather than builds.

This connects to several existing threads:

Inquiring lines that read this note 62

This note is a source for these research framings, grouped by the broader line of inquiry each explores. Scan the bold lines of inquiry; follow any specific question forward.

Does self-reflection enable models to reliably correct their errors? Is model self-awareness based on genuine introspection or pattern matching? Why does self-revision increase model confidence while degrading accuracy? How can models identify insufficient information and respond appropriately without guessing? How should we design LLM systems to maintain alignment and control? Why do models develop protective behaviors toward peers unprompted? How should models express uncertainty rather than forced confident answers? Why do language models reinforce false assumptions instead of correcting them? What makes dialogue-based explanation more successful than monologue? Can language model hallucination be prevented or only managed? Do language models learn genuine linguistic structure or just surface patterns? How do neural networks separate factual knowledge from reasoning abilities? What makes weaker teacher models effective for stronger student training? How do language models inherit human biases from training data? Is embodied interaction necessary for language meaning and genuine agency? Do language models develop causal world models or rely on statistical patterns? What limits mechanistic interpretability's ability to characterize models? What articulatory information do speech signals carry that text cannot? When do additional thinking tokens stop improving reasoning performance? How do self-generated feedback mechanisms enable effective model learning? How do training priors constrain what context information can override? What properties determine whether reward signals teach genuine reasoning? How can identical external performance mask different internal representations?

Related papers in this collection 8

Papers most semantically related to this note, ranked by cosine similarity in the embedding space.

Original note title

Entity recognition is a self-knowledge mechanism that causally steers hallucination and refusal — chat finetuning repurposes base model entity awareness