Towards Fully Exploiting LLM Internal States to Enhance Knowledge Boundary Perception

Large language models (LLMs) exhibit impressive performance across diverse tasks but often struggle to accurately gauge their knowledge boundaries, leading to confident yet incorrect responses. This paper explores leveraging LLMs' internal states to enhance their perception of knowledge boundaries from efficiency and risk perspectives. We investigate whether LLMs can estimate their confidence using internal states before response generation, potentially saving computational resources. Our experiments on datasets like Natural Questions, HotpotQA, and MMLU reveal that LLMs demonstrate significant pre-generation perception, which is further refined post-generation, with perception gaps remaining stable across varying conditions. To mitigate risks in critical domains, we introduce Consistency-based Confidence Calibration ($C^3$), which assesses confidence consistency through question reformulation. $C^3$ significantly improves LLMs' ability to recognize their knowledge gaps, enhancing the unknown perception rate by 5.6\% on NQ and 4.9\% on HotpotQA. Our findings suggest that pre-generation confidence estimation can optimize efficiency, while $C^3$ effectively controls output risks, advancing the reliability of LLMs in practical applications.

Are Large Language Models More Honest in Their Probabilistic or Verbalized Confidence?

Large language models (LLMs) have been found to produce hallucinations when the question exceeds their internal knowledge boundaries. A reliable model should have a clear perception of its knowledge boundaries, providing correct answers within its scope and refusing to answer when it lacks knowledge. Existing research on LLMs' perception of their knowledge boundaries typically uses either the probability of the generated tokens or the verbalized confidence as the model's confidence in its response. However, these studies overlook the differences and connections between the two. In this paper, we conduct a comprehensive analysis and comparison of LLMs' probabilistic perception and verbalized perception of their factual knowledge boundaries. First, we investigate the pros and cons of these two perceptions. Then, we study how they change under questions of varying frequencies. Finally, we measure the correlation between LLMs' probabilistic confidence and verbalized confidence. Experimental results show that 1) LLMs' probabilistic perception is generally more accurate than verbalized perception but requires an in-domain validation set to adjust the confidence threshold. 2) Both perceptions perform better on less frequent questions. 3) It is challenging for LLMs to accurately express their internal confidence in natural language.

Contextual Dual Learning Algorithm with Listwise Distillation for Unbiased Learning to Rank

Unbiased Learning to Rank (ULTR) aims to leverage biased implicit user feedback (e.g., click) to optimize an unbiased ranking model. The effectiveness of the existing ULTR methods has primarily been validated on synthetic datasets. However, their performance on real-world click data remains unclear. Recently, Baidu released a large publicly available dataset of their web search logs. Subsequently, the NTCIR-17 ULTRE-2 task released a subset dataset extracted from it. We conduct experiments on commonly used or effective ULTR methods on this subset to determine whether they maintain their effectiveness. In this paper, we propose a Contextual Dual Learning Algorithm with Listwise Distillation (CDLA-LD) to simultaneously address both position bias and contextual bias. We utilize a listwise-input ranking model to obtain reconstructed feature vectors incorporating local contextual information and employ the Dual Learning Algorithm (DLA) method to jointly train this ranking model and a propensity model to address position bias. As this ranking model learns the interaction information within the documents list of the training set, to enhance the ranking model's generalization ability, we additionally train a pointwise-input ranking model to learn the listwise-input ranking model's capability for relevance judgment in a listwise manner. Extensive experiments and analysis confirm the effectiveness of our approach.

When Do LLMs Need Retrieval Augmentation? Mitigating LLMs' Overconfidence Helps Retrieval Augmentation

Large Language Models (LLMs) have been found to have difficulty knowing they do not possess certain knowledge and tend to provide specious answers in such cases. Retrieval Augmentation (RA) has been extensively studied to mitigate LLMs' hallucinations. However, due to the extra overhead and unassured quality of retrieval, it may not be optimal to conduct RA all the time. A straightforward idea is to only conduct retrieval when LLMs are uncertain about a question. This motivates us to enhance the LLMs' ability to perceive their knowledge boundaries to help RA. In this paper, we first quantitatively measure LLMs' such ability and confirm their overconfidence. Then, we study how LLMs' certainty about a question correlates with their dependence on external retrieved information. We propose several methods to enhance LLMs' perception of knowledge boundaries and show that they are effective in reducing overconfidence. Additionally, equipped with these methods, LLMs can achieve comparable or even better performance of RA with much fewer retrieval calls.

A Comparative Study of Training Objectives for Clarification Facet Generation

Due to the ambiguity and vagueness of a user query, it is essential to identify the query facets for the clarification of user intents. Existing work on query facet generation has achieved compelling performance by sequentially predicting the next facet given previously generated facets based on pre-trained language generation models such as BART. Given a query, there are mainly two types of training objectives to guide the facet generation models. One is to generate the default sequence of ground-truth facets, and the other is to enumerate all the permutations of ground-truth facets and use the sequence that has the minimum loss for model updates. The second is permutation-invariant while the first is not. In this paper, we aim to conduct a systematic comparative study of various types of training objectives, with different properties of not only whether it is permutation-invariant but also whether it conducts sequential prediction and whether it can control the count of output facets. To this end, we propose another three training objectives of different aforementioned properties. For comprehensive comparisons, besides the commonly used evaluation that measures the matching with ground-truth facets, we also introduce two diversity metrics to measure the diversity of the generated facets. Based on an open-domain query facet dataset, i.e., MIMICS, we conduct extensive analyses and show the pros and cons of each method, which could shed light on model training for clarification facet generation. The code can be found at \url{https://github.com/ShiyuNee/Facet-Generation}