Ponhvoan Srey

Ponhvoan Srey, Xiaobao Wu, Cong-Duy Nguyen et al.

Uncertainty estimation is a promising approach to detect hallucinations in large language models (LLMs). Recent approaches commonly depend on model internal states to estimate uncertainty. However, they suffer from strict assumptions on how hidden states should evolve across layers, and from information loss by solely focusing on last or mean tokens. To address these issues, we present Sequential Internal Variance Representation (SIVR), a supervised hallucination detection framework that leverages token-wise, layer-wise features derived from hidden states. SIVR adopts a more basic assumption that uncertainty manifests in the degree of dispersion or variance of internal representations across layers, rather than relying on specific assumptions, which makes the method model and task agnostic. It additionally aggregates the full sequence of per-token variance features, learning temporal patterns indicative of factual errors and thereby preventing information loss. Experimental results demonstrate SIVR consistently outperforms strong baselines. Most importantly, SIVR enjoys stronger generalisation and avoids relying on large training sets, highlighting the potential for practical deployment. Our code repository is available online at https://github.com/ponhvoan/internal-variance.

Ponhvoan Srey, Quang Minh Nguyen, Xiaobao Wu et al.

Uncertainty estimation (UE) aims to detect hallucinated outputs of large language models (LLMs) to improve their reliability. However, UE metrics often exhibit unstable performance across configurations, which significantly limits their applicability. In this work, we formalise this phenomenon as proxy failure, since most UE metrics originate from model behaviour, rather than being explicitly grounded in the factual correctness of LLM outputs. With this, we show that UE metrics become non-discriminative precisely in low-information regimes. To alleviate this, we propose Truth AnChoring (TAC), a post-hoc calibration method to remedy UE metrics, by mapping the raw scores to truth-aligned scores. Even with noisy and few-shot supervision, our TAC can support the learning of well-calibrated uncertainty estimates, and presents a practical calibration protocol. Our findings highlight the limitations of treating heuristic UE metrics as direct indicators of truth uncertainty, and position our TAC as a necessary step toward more reliable uncertainty estimation for LLMs. The code repository is available at https://github.com/ponhvoan/TruthAnchor/.

Ponhvoan Srey, Yaxin Shi, Hangwei Qian et al.

Fully Test-Time Adaptation (FTTA) addresses domain shifts without access to source data and training protocols of the pre-trained models. Traditional strategies that align source and target feature distributions are infeasible in FTTA due to the absence of training data and unpredictable target domains. In this work, we exploit a dual perspective on FTTA, and propose Agnostic FTTA (AFTTA) as a novel formulation that enables the usage of off-the-shelf domain transformations during test-time to enable direct generalization to unforeseeable target data. To address this, we develop an uncover-and-unlearn approach. First, we uncover potential unwanted shifts between source and target domains by simulating them through predefined mappings and consider them as nuisances. Then, during test-time prediction, the model is enforced to unlearn these nuisances by regularizing the consequent shifts in latent representations and label predictions. Specifically, a mutual information-based criterion is devised and applied to guide nuisances unlearning in the feature space and encourage confident and consistent prediction in label space. Our proposed approach explicitly addresses agnostic domain shifts, enabling superior model generalization under FTTA constraints. Extensive experiments on various tasks, involving corruption and style shifts, demonstrate that our method consistently outperforms existing approaches.

Ponhvoan Srey, Xiaobao Wu, Anh Tuan Luu

Unsupervised hallucination detection aims to identify hallucinated content generated by large language models (LLMs) without relying on labeled data. While unsupervised methods have gained popularity by eliminating labor-intensive human annotations, they frequently rely on proxy signals unrelated to factual correctness. This misalignment biases detection probes toward superficial or non-truth-related aspects, limiting generalizability across datasets and scenarios. To overcome these limitations, we propose IRIS, an unsupervised hallucination detection framework, leveraging internal representations intrinsic to factual correctness. IRIS prompts the LLM to carefully verify the truthfulness of a given statement, and obtain its contextualized embedding as informative features for training. Meanwhile, the uncertainty of each response is considered a soft pseudolabel for truthfulness. Experimental results demonstrate that IRIS consistently outperforms existing unsupervised methods. Our approach is fully unsupervised, computationally low cost, and works well even with few training data, making it suitable for real-time detection.