Does Bielik Know What It Doesn't Know? Activation Dispersion Separates Entity Familiarity from Factual Reliability Across Model Scale
Quick Answer
The study investigates how activation dispersion in Bielik models (1.5B-11B parameters) distinguishes entity familiarity from factual reliability, achieving AUROC scores up to 1.00 across various domains.
Quick Take
The study investigates how activation dispersion in Bielik models (1.5B-11B parameters) distinguishes entity familiarity from factual reliability, achieving AUROC scores up to 1.00 across various domains. Despite high entity recognition, models struggle with factual accuracy, showing a stark contrast in performance as model size increases.
Key Points
- Four Bielik models were tested across 504 prompts in four entity domains.
- Dispersion measures achieved AUROC scores between 0.95 and 1.00 for known vs. fabricated entities.
- Models exhibited high entity familiarity but low factual reliability, especially for known entities.
- A five-sample semantic-entropy baseline reached only 0.71-0.83 at higher inference costs.
- Models rarely abstain from answering, with only 2 refusals in 2,520 audits.
Paper Resources
📖 Reader Mode
~2 min readAbstract:Large language models hallucinate most about entities they have never seen. We ask whether a model's activations betray entity familiarity before a single answer token is generated, and whether that signal predicts the factual reliability of the answers. On four Polish Bielik models (1.5B-11B parameters), we probe four entity domains (athletes, cities, writers, musicians), each with 42 well-known, 42 obscure-but-real, and 42 fabricated entities addressed by a one-sentence question (504 prompts per model). Two unsupervised, single-forward-pass dispersion measures over post-SwiGLU MLP activations, inverse participation ratio and spectral entropy, separate known from fabricated entities at AUROC 0.95-1.00 across all domains and scales; a supervised linear probe reaches 0.99-1.00. Both clear selection-aware permutation floors of about 0.70-0.74 (empirical p<=1e-3), survive held-out layer selection (0.93-0.99), and persist on real names (known vs. obscure-but-real: 0.96-1.00). The signal transfers across entity types (mean off-diagonal AUROC 0.92-0.99); a matched-template counterfactual shows the only large drops are template-caused, not entity-type effects, and the signal is diffuse across heads. This representational signal is already at ceiling at 1.5B, whereas behavioral factual reliability scales sharply: 0, 2, 10, and 19 of 42 known athletes are answered fully correctly by the 1.5B, 4.5B, 7B, and 11B models under a strict judge. Within known entities, separating correct from hallucinated answers is much harder (probe 0.93; dispersion no better than a first-token-entropy baseline). A five-sample semantic-entropy baseline reaches only 0.71-0.83 at 5x the inference cost. Despite this internal awareness, the models almost never abstain: an audit of 2,520 answers finds 2 refusals and 1 hedge. Entity familiarity and factual reliability are distinct phenomena on different scaling curves.
| Comments: | 23 pages, 6 figures and 7 tables |
| Subjects: | Computation and Language (cs.CL); Machine Learning (cs.LG) |
| Cite as: | arXiv:2607.07670 [cs.CL] |
| (or arXiv:2607.07670v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2607.07670 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Grzegorz Brzezinka [view email]
[v1]
Wed, 8 Jul 2026 17:24:43 UTC (157 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.CL
See more →Quantifying Prior Dominance in Systems
The study introduces the Normalized Context Utilization (NCU) metric to evaluate Retrieval-Augmented Generation (RAG) systems, revealing that Small Language Models (SLMs) outperform larger models in factual extraction. The findings indicate that traditional scaling laws yield diminishing returns, with a commercial API frequently failing against adversarial evidence due to systemic confidence collapse.