Learned Readout Tokens for Extracting Large Language Model Confidence

Nima Seifi; Martin Ester

cs.CLOct 20, 2026

Learned Readout Tokens for Extracting Large Language Model Confidence

Authors: Nima Seifi, Martin Ester

Organizations: School of Computing Science, Simon Fraser University, 8888 University Drive, Burnaby, BC V5A 1S6, Canada

Abstract

Large language models (LLMs) are increasingly used within everyday applications where their responses have significant downstream impacts. For responsible deployment of such systems, we desire insight into how much we can trust their individual outputs. A popular approach for estimating LLM confidence is to train probes on the hidden states of the LLM's tokens to predict the correctness of the model's output. These probes typically read hidden states from tokens that are already in the model's vocabulary. In this work, we introduce a new approach to confidence probes that learn a new readout token and utilize its hidden states to train the probe. We show that our learned readout approach ScratchProbe outperforms its existing-token counterpart on a variety of datasets and LLMs. We also extend our method to a multi-token approach which we show provides further performance improvements. Finally, since the learned readout changes only the position a probe reads from, any existing probe can adopt it. Applied to a state-of-the-art probe, it improves AUROC on every backbone, in domain and out of domain.

Figures & tables

Figure 1: (a) The standard last-token probe from prior work and (b) ScratchProbe . Both use the same frozen LLM and MLP probe; only the probed position differs.

	Llama-3.2-3B		Llama-3.1-8B		Phi-4-mini (3.8B)
Method	ID	OOD	ID	OOD	ID	OOD
Seq. likelihood	0.816	0.688	0.829	0.698	0.771	0.658
P(True)	0.727	0.594	0.719	0.608	0.783	0.714
Verbalized conf.	0.744	0.613	0.714	0.645	0.553	0.588
Self-consistency ( $k{=}10$ )	0.873	0.737	0.873	0.745	0.835	0.734
Semantic entropy ( $k{=}10$ )	0.840	0.723	0.844	0.730	0.806	0.718

Table 1: Macro-average AUROC (

\uparrow

) of ScratchProbe and baselines on three frozen instruction-tuned backbones. Bold marks the best and underline marks the second-best value in each column. Per-dataset results appear in Tables S1 – S3 of the technical supplement.

	Llama-3.2-3B		Llama-3.1-8B		Phi-4-mini (3.8B)
Method	ID	OOD	ID	OOD	ID	OOD
Seq. likelihood	0.093	0.261	0.073	0.267	0.140	0.288
P(True)	0.378	0.298	0.221	0.369	0.149	0.207
Verbalized conf.	0.107	0.183	0.098	0.308	0.278	0.407
Self-consistency ( $k{=}10$ )	0.072	0.151	0.064	0.154	0.071	0.161
Semantic entropy ( $k{=}10$ )	0.123	0.120	0.105	0.122	0.124	0.134

Table 2: Macro-average ECE (

\downarrow

, 15 equal-mass bins). Bold marks the best and underline marks the second-best value in each column. Per-dataset results appear in Tables S4 – S6 of the technical supplement.

Input-Output Example
Question: Divided government frequently results in: A) reapportionment of voting districts; B) amendments to the Constitution; C) conflict between states and the federal government; D) delays in the confirmation of those nominated to be ambassadors.
Model Answer: C Ground Truth: D Grader: Incorrect
Probe	Confidence
Last-token	0.81
ScratchProbe ( $n{=}8$ )	0.17

Table 3: Input and output example. The last-token probe is confident in a wrong answer; ScratchProbe is not.

Figure 2: Macro AUROC on Phi-4-mini against

n

. The dashed line is the last-token probe.

Figure 3: Macro ECE on Phi-4-mini against

n

. The dashed line is the last-token probe.

	AUROC ( $\uparrow$ )						ECE ( $\downarrow$ )
	Llama-3.2-3B		Llama-3.1-8B		Phi-4-mini		Llama-3.2-3B		Llama-3.1-8B		Phi-4-mini
Probe	ID	OOD	ID	OOD	ID	OOD	ID	OOD	ID	OOD	ID	OOD
I 2 (last answer token)	0.872	0.723	0.895	0.730	0.927	0.714	0.065	0.157	0.043	0.130	0.037	0.146
I 2 + ScratchProbe tokens ( $n{=}8$ )	0.900	0.754	0.906	0.742	0.935	0.757	0.055	0.137	0.039	0.112	0.045	0.173

Table 4: Macro-average AUROC (

\uparrow

) and ECE (

\downarrow

) of I 2 with its original last-token readout compared against its ScratchProbe variant.

Figure 4: Selective accuracy against coverage on Phi-4-mini. At each coverage threshold, the most confident fraction of answers is kept and the accuracy of the kept set is reported.

Appendix figures & tables7 assets

Supplementary material from the paper’s appendix.

Appendix

	In-domain			Out-of-domain				Macro avg.
Method	TriviaQA	SciQ	GSM8K	MMLU	TruthfulQA	CSQA	MATH-500	ID	OOD
Seq. likelihood	0.839	0.893	0.716	0.750	0.704	0.759	0.537	0.816	0.688
P(True)	0.800	0.670	0.712	0.537	0.575	0.586	0.677	0.727	0.594
Verbalized conf.	0.799	0.760	0.672	0.604	0.581	0.563	0.704	0.744	0.613
Self-consistency ( $k{=}10$ )	0.862	0.796	0.962	0.722	0.606	0.719	0.900	0.873	0.737
Semantic entropy ( $k{=}10$ )	0.860	0.794	0.867	0.717	0.632	0.718	0.825	0.840	0.723

Appendix

Table S1: AUROC (

\uparrow

) on Llama-3.2-3B-Instruct, per dataset. Probe rows report mean

\pm

std over three training seeds. Sequence likelihood, P(True), and verbalized confidence are averaged over the same three seeds and have negligible spread. Self-consistency and semantic entropy use

k{=}10

samples per question in a single run. Bold marks the best and underline the second-best value in each column.

	In-domain			Out-of-domain				Macro avg.
Method	TriviaQA	SciQ	GSM8K	MMLU	TruthfulQA	CSQA	MATH-500	ID	OOD
Seq. likelihood	0.862	0.903	0.721	0.784	0.660	0.760	0.588	0.829	0.698
P(True)	0.854	0.569	0.734	0.560	0.525	0.569	0.778	0.719	0.608
Verbalized conf.	0.788	0.762	0.591	0.651	0.561	0.648	0.717	0.714	0.645
Self-consistency ( $k{=}10$ )	0.881	0.800	0.940	0.766	0.621	0.695	0.899	0.873	0.745
Semantic entropy ( $k{=}10$ )	0.878	0.799	0.855	0.760	0.643	0.694	0.824	0.844	0.730

Appendix

Table S2: AUROC (

\uparrow

) on Llama-3.1-8B-Instruct, per dataset. Probe rows report mean

\pm

std over three training seeds. Sequence likelihood, P(True), and verbalized confidence are averaged over the same three seeds and have negligible spread. Self-consistency and semantic entropy use

k{=}10

samples per question in a single run. Bold marks the best and underline the second-best value in each column.

	In-domain			Out-of-domain				Macro avg.
Method	TriviaQA	SciQ	GSM8K	MMLU	TruthfulQA	CSQA	MATH-500	ID	OOD
Seq. likelihood	0.844	0.820	0.651	0.684	0.665	0.737	0.545	0.771	0.658
P(True)	0.844	0.891	0.614	0.687	0.685	0.682	0.802	0.783	0.714
Verbalized conf.	0.566	0.583	0.510	0.545	0.644	0.507	0.657	0.553	0.588
Self-consistency ( $k{=}10$ )	0.854	0.796	0.856	0.709	0.602	0.732	0.893	0.835	0.734
Semantic entropy ( $k{=}10$ )	0.853	0.791	0.774	0.706	0.643	0.727	0.798	0.806	0.718

Appendix

Table S3: AUROC (

\uparrow

) on Phi-4-mini-instruct (3.8B), per dataset. Probe rows report mean

\pm

std over three training seeds. Sequence likelihood, P(True), and verbalized confidence are averaged over the same three seeds and have negligible spread. Self-consistency and semantic entropy use

k{=}10

samples per question in a single run. Bold marks the best and underline the second-best value in each column.

	In-domain			Out-of-domain				Macro avg.
Method	TriviaQA	SciQ	GSM8K	MMLU	TruthfulQA	CSQA	MATH-500	ID	OOD
Seq. likelihood	0.147	0.038	0.095	0.100	0.365	0.110	0.468	0.093	0.261
P(True)	0.167	0.407	0.560	0.337	0.267	0.350	0.239	0.378	0.298
Verbalized conf.	0.134	0.158	0.031	0.082	0.302	0.094	0.255	0.107	0.183
Self-consistency ( $k{=}10$ )	0.063	0.043	0.110	0.125	0.252	0.138	0.089	0.072	0.151
Semantic entropy ( $k{=}10$ )	0.118	0.063	0.188	0.069	0.189	0.099	0.125	0.123	0.120

Appendix

Table S4: ECE (

\downarrow

) on Llama-3.2-3B-Instruct, per dataset. ECE uses 15 equal-mass bins. Probe rows report mean

\pm

std over three training seeds. Sequence likelihood, P(True), and verbalized confidence are averaged over the same three seeds and have negligible spread. Self-consistency and semantic entropy use

k{=}10

samples per question in a single run. Bold marks the best and underline the second-best value in each column.

	In-domain			Out-of-domain				Macro avg.
Method	TriviaQA	SciQ	GSM8K	MMLU	TruthfulQA	CSQA	MATH-500	ID	OOD
Seq. likelihood	0.115	0.042	0.062	0.116	0.366	0.140	0.448	0.073	0.267
P(True)	0.157	0.378	0.128	0.277	0.439	0.272	0.489	0.221	0.369
Verbalized conf.	0.168	0.033	0.093	0.207	0.459	0.099	0.466	0.098	0.308
Self-consistency ( $k{=}10$ )	0.047	0.053	0.092	0.137	0.241	0.161	0.077	0.064	0.154
Semantic entropy ( $k{=}10$ )	0.118	0.036	0.162	0.087	0.173	0.123	0.103	0.105	0.122

Appendix

Table S5: ECE (

\downarrow

) on Llama-3.1-8B-Instruct, per dataset. ECE uses 15 equal-mass bins. Probe rows report mean

\pm

std over three training seeds. Sequence likelihood, P(True), and verbalized confidence are averaged over the same three seeds and have negligible spread. Self-consistency and semantic entropy use

k{=}10

samples per question in a single run. Bold marks the best and underline the second-best value in each column.

	In-domain			Out-of-domain				Macro avg.
Method	TriviaQA	SciQ	GSM8K	MMLU	TruthfulQA	CSQA	MATH-500	ID	OOD
Seq. likelihood	0.200	0.095	0.124	0.300	0.278	0.190	0.384	0.140	0.288
P(True)	0.178	0.047	0.221	0.157	0.351	0.109	0.210	0.149	0.207
Verbalized conf.	0.447	0.115	0.272	0.329	0.565	0.279	0.454	0.278	0.407
Self-consistency ( $k{=}10$ )	0.064	0.064	0.086	0.172	0.236	0.110	0.125	0.071	0.161
Semantic entropy ( $k{=}10$ )	0.115	0.054	0.203	0.114	0.174	0.085	0.163	0.124	0.134

Appendix

Table S6: ECE (

\downarrow

) on Phi-4-mini-instruct (3.8B), per dataset. ECE uses 15 equal-mass bins. Probe rows report mean

\pm

std over three training seeds. Sequence likelihood, P(True), and verbalized confidence are averaged over the same three seeds and have negligible spread. Self-consistency and semantic entropy use

k{=}10

samples per question in a single run. Bold marks the best and underline the second-best value in each column.

AUROC ( $\uparrow$ )
	In-domain			Out-of-domain				Macro avg.
Probe	TriviaQA	SciQ	GSM8K	MMLU	TruthfulQA	CSQA	MATH-500	ID	OOD
Llama-3.2-3B-Instruct
I 2 (last answer token)	0.893 $\pm$ 0.002	0.886 $\pm$ 0.011	0.837 $\pm$ 0.006	0.737 $\pm$ 0.008	0.603 $\pm$ 0.013	0.725 $\pm$ 0.016	0.828 $\pm$ 0.020	0.872 $\pm$ 0.002	0.723 $\pm$ 0.003
I 2 + ScratchProbe tokens ( $n{=}8$ )	0.902	0.907	0.891	0.749	0.627	0.764	0.877	0.900	0.754
Llama-3.1-8B-Instruct

Appendix

Table S7: Per-dataset AUROC (

\uparrow

) and ECE (

\downarrow

) of I 2 with its original last-answer-token readout and of the ScratchProbe variant that reads from <CONF> after eight jointly trained scratch tokens. The I 2 rows report mean

\pm

std over three encoder seeds. The ScratchProbe variant rows are from a single seed (42). Bold marks the better of the two rows in each column.

Confidence Estimation in Language Models

Confidence estimation in language models measures whether a model’s uncertainty signals predict answer correctness and support reliable decisions.

See all 249 papers

Calibrating Overconfidence Without Sacrificing Confidence: Probe-Conditioned Head Intervention for LLMs
Jun 2, 2026Ke Li, Chongzhe Zhang, Zifan Zeng +3LLM Reliability Causal Interventions in Language Models
Huawei Heisenberg Research Center · EPFL · TU Berlin +1
How do LLMs Compute Verbal Confidence
Mar 18, 2026Dharshan Kumaran, Arthur Conmy, Federico Barbero +3Verbalized Confidence Estimation Confidence Estimation in Language Models
Google DeepMind
Know When You're Wrong: Aligning Confidence with Correctness for LLM Error Detection
Feb 18, 2026Xie Xiaohu, Liu Xiaohu, Yao BenjaminLanguage Model Error Detection Confidence Estimation in Language Models
Alexa AI, Amazon, Seattle, WA, USA

Jun 2, 2026J/K move · Enter open · S save

Calibrating Overconfidence Without Sacrificing Confidence: Probe-Conditioned Head Intervention for LLMs

Ke Li, Chongzhe Zhang, Zifan Zeng, Feng Liu+2

Huawei Heisenberg Research Center · EPFL · TU Berlin+1

LLM Reliability Causal Interventions in Language Models Language Model Calibration Confidence Estimation in Language Models LLM Uncertainty Estimation Verbalized Confidence Estimation+1

Open paper AskPDF arXiv

Large language models often express high confidence in answers that are wrong. Standard calibration remedies typically act globally or at the score level, reducing unwarranted confidence but also risking erosion of warranted confidence on correct answers. We introduce Probe-Conditioned Head Intervention (PCHI), an inference-time method that uses a frozen probe to detect likely wrong-but-confident responses and conditionally rescales downstream attention-head outputs during confidence generation. On Qwen3-4B-Instruct solving OpenMathInstruct problems with a structured binary confidence field, readout-token PCHI converts 82.2% of originally wrong-yes confidence readouts to

\texttt{no}

, while a joint intervention across upstream confidence-template tokens reduces ECE from 21.9% to 9.2% and damages only 5.1% of originally correct-yes readouts. The readout-token effect also appears on Gemma3-4B, though upstream interventions are weaker and more mask-dependent. These results show that verbalized overconfidence can be selectively reduced through conditionally applied internal intervention, partially decoupling the suppression of unwarranted confidence from the loss of warranted confidence.

Abstract

Confidence Estimation in Language Models

Explore similar work

Calibrating Overconfidence Without Sacrificing Confidence: Probe-Conditioned Head Intervention for LLMs

How do LLMs Compute Verbal Confidence

Know When You're Wrong: Aligning Confidence with Correctness for LLM Error Detection