[2509.10625] No Answer Needed: Predicting LLM Answer Accuracy from Question-Only Linear Probes

Computer Science > Computation and Language arXiv:2509.10625 (cs) [Submitted on 12 Sep 2025 (v1), last revised 3 Mar 2026 (this version, v3)] Title:No Answer Needed: Predicting LLM Answer Accuracy from Question-Only Linear Probes Authors:Iván Vicente Moreno Cencerrado, Arnau Padrés Masdemont, Anton Gonzalvez Hawthorne, David Demitri Africa, Lorenzo Pacchiardi View a PDF of the paper titled No Answer Needed: Predicting LLM Answer Accuracy from Question-Only Linear Probes, by Iv\'an Vicente Moreno Cencerrado and 4 other authors View PDF HTML (experimental) Abstract:Do large language models (LLMs) anticipate when they will answer correctly? To study this, we extract activations after a question is read but before any tokens are generated, and train linear probes to predict whether the model's forthcoming answer will be correct. Across three open-source model families ranging from 7 to 70 billion parameters, projections on this "in-advance correctness direction" trained on generic trivia questions predict success in distribution and on diverse out-of-distribution knowledge datasets, indicating a deeper signal than dataset-specific spurious features, and outperforming black-box baselines and verbalised predicted confidence. Predictive power saturates in intermediate layers and, notably, generalisation falters on questions requiring mathematical reasoning. Moreover, for models responding "I don't know", doing so strongly correlates with the probe score, indicating that the same ...