AI Insight
Researchers investigated whether language models can assess their own familiarity with entities before generating answers by analyzing neural activations in twelve instruction-tuned models using a dataset of 1,440 Polish entities plus fabricated controls. They found that "familiarity probes" successfully distinguish real from fake entities across all model families, with Polish-adapted models (Bielik and PLLuM) showing stronger correlation with entity popularity than multilingual models. The familiarity signal remains robust across prompt languages and can be used to steer model refusal behavior, moving refusal rates from 0.24 to 1.00 for known entities and from 0.73 to 0.00 for unknown ones in controlled experiments.
Why it matters
This work demonstrates that language models possess internal representations of entity familiarity that could be leveraged to reduce hallucinations by allowing models to refuse answering questions about unfamiliar topics. The findings suggest practical methods for improving AI reliability through pre-generation uncertainty detection, which could be particularly valuable for high-stakes applications requiring factual accuracy.
Understand the Science
arXiv:2607.13568v1 Announce Type: cross
Abstract: Can a language model estimate its familiarity with an entity before generating an answer? We study activations at the final prompt token in twelve instruction-tuned models from the Bielik, PLLuM, Gemma-4, and Qwen3 families, using a new dataset of 1,440 Polish entities spanning four domains and ten Wikipedia-pageview deciles, plus fabricated controls. Familiarity-probe scores separate real from fabricated entities in every family; in the Polish-adapted Bielik and PLLuM families they additionally track entity popularity (model-mean Spearman $rho$ 0.28-0.57, versus at most 0.11 in Gemma-4 and Qwen3), a pattern more strongly associated with Polish adaptation than with parameter count in this model sample. In a paired experiment on two families, probes retain 96-101% of within-language AUROC when the Polish question stem is replaced with an English one around unchanged entity names, showing robustness to prompt language in this setting. In Gemma-4-12B, the only model that natively refuses, adding a one-dimensional familiarity direction at a single layer moves refusal rates monotonically in both directions (0.24 to 1.00 on well-known entities; 0.73 to 0.00 on unknown ones). Finally, a calibrated familiarity probe is competitive among pre-generation abstention gates, although post-generation detectors better predict behavioral error on average. These results support a graded pre-generation entity-familiarity readout, and a separation between representational familiarity and the policy that converts it into abstention.