Social Bias Evaluation for Large Language Models Requires Prompt Variations

Warning: This paper contains examples of stereotypes and biases. Large Language Models (LLMs) exhibit considerable social biases, and various studies have tried to evaluate and mitigate these biases accurately. Previous studies use downstream tasks as prompts to examine the degree of social biases for evaluation and mitigation. While LLMs' output highly depends on prompts, previous studies evaluating and mitigating bias have often relied on a limited variety of prompts. In this paper, we investigate the sensitivity of LLMs when changing prompt variations (task instruction and prompt, few-shot examples, debias-prompt) by analyzing task performance and social bias of LLMs. Our experimental results reveal that LLMs are highly sensitive to prompts to the extent that the ranking of LLMs fluctuates when comparing models for task performance and social bias. Additionally, we show that LLMs have tradeoffs between performance and social bias caused by the prompts. Less bias from prompt setting may result in reduced performance. Moreover, the ambiguity of instances is one of the reasons for this sensitivity to prompts in advanced LLMs, leading to various outputs. We recommend using diverse prompts, as in this study, to compare the effects of prompts on social bias in LLMs.

Sampling-based Pseudo-Likelihood for Membership Inference Attacks

Large Language Models (LLMs) are trained on large-scale web data, which makes it difficult to grasp the contribution of each text. This poses the risk of leaking inappropriate data such as benchmarks, personal information, and copyrighted texts in the training data. Membership Inference Attacks (MIA), which determine whether a given text is included in the model's training data, have been attracting attention. Previous studies of MIAs revealed that likelihood-based classification is effective for detecting leaks in LLMs. However, the existing methods cannot be applied to some proprietary models like ChatGPT or Claude 3 because the likelihood is unavailable to the user. In this study, we propose a Sampling-based Pseudo-Likelihood (\textbf{SPL}) method for MIA (\textbf{SaMIA}) that calculates SPL using only the text generated by an LLM to detect leaks. The SaMIA treats the target text as the reference text and multiple outputs from the LLM as text samples, calculates the degree of $n$-gram match as SPL, and determines the membership of the text in the training data. Even without likelihoods, SaMIA performed on par with existing likelihood-based methods.

A Little Leak Will Sink a Great Ship: Survey of Transparency for Large Language Models from Start to Finish

Large Language Models (LLMs) are trained on massive web-crawled corpora. This poses risks of leakage, including personal information, copyrighted texts, and benchmark datasets. Such leakage leads to undermining human trust in AI due to potential unauthorized generation of content or overestimation of performance. We establish the following three criteria concerning the leakage issues: (1) leakage rate: the proportion of leaked data in training data, (2) output rate: the ease of generating leaked data, and (3) detection rate: the detection performance of leaked versus non-leaked data. Despite the leakage rate being the origin of data leakage issues, it is not understood how it affects the output rate and detection rate. In this paper, we conduct an experimental survey to elucidate the relationship between the leakage rate and both the output rate and detection rate for personal information, copyrighted texts, and benchmark data. Additionally, we propose a self-detection approach that uses few-shot learning in which LLMs detect whether instances are present or absent in their training data, in contrast to previous methods that do not employ explicit learning. To explore the ease of generating leaked information, we create a dataset of prompts designed to elicit personal information, copyrighted text, and benchmarks from LLMs. Our experiments reveal that LLMs produce leaked information in most cases despite less such data in their training set. This indicates even small amounts of leaked data can greatly affect outputs. Our self-detection method showed superior performance compared to existing detection methods.

Likelihood-based Mitigation of Evaluation Bias in Large Language Models

Large Language Models (LLMs) are widely used to evaluate natural language generation tasks as automated metrics. However, the likelihood, a measure of LLM's plausibility for a sentence, can vary due to superficial differences in sentences, such as word order and sentence structure. It is therefore possible that there might be a likelihood bias if LLMs are used for evaluation: they might overrate sentences with higher likelihoods while underrating those with lower likelihoods. In this paper, we investigate the presence and impact of likelihood bias in LLM-based evaluators. We also propose a method to mitigate the likelihood bias. Our method utilizes highly biased instances as few-shot examples for in-context learning. Our experiments in evaluating the data-to-text and grammatical error correction tasks reveal that several LLMs we test display a likelihood bias. Furthermore, our proposed method successfully mitigates this bias, also improving evaluation performance (in terms of correlation of models with human scores) significantly.

Eagle: Ethical Dataset Given from Real Interactions

Recent studies have demonstrated that large language models (LLMs) have ethical-related problems such as social biases, lack of moral reasoning, and generation of offensive content. The existing evaluation metrics and methods to address these ethical challenges use datasets intentionally created by instructing humans to create instances including ethical problems. Therefore, the data does not reflect prompts that users actually provide when utilizing LLM services in everyday contexts. This may not lead to the development of safe LLMs that can address ethical challenges arising in real-world applications. In this paper, we create Eagle datasets extracted from real interactions between ChatGPT and users that exhibit social biases, toxicity, and immoral problems. Our experiments show that Eagle captures complementary aspects, not covered by existing datasets proposed for evaluation and mitigation of such ethical challenges. Our code is publicly available at https://huggingface.co/datasets/MasahiroKaneko/eagle.

Evaluating Gender Bias in Large Language Models via Chain-of-Thought Prompting

There exist both scalable tasks, like reading comprehension and fact-checking, where model performance improves with model size, and unscalable tasks, like arithmetic reasoning and symbolic reasoning, where model performance does not necessarily improve with model size. Large language models (LLMs) equipped with Chain-of-Thought (CoT) prompting are able to make accurate incremental predictions even on unscalable tasks. Unfortunately, despite their exceptional reasoning abilities, LLMs tend to internalize and reproduce discriminatory societal biases. Whether CoT can provide discriminatory or egalitarian rationalizations for the implicit information in unscalable tasks remains an open question. In this study, we examine the impact of LLMs' step-by-step predictions on gender bias in unscalable tasks. For this purpose, we construct a benchmark for an unscalable task where the LLM is given a list of words comprising feminine, masculine, and gendered occupational words, and is required to count the number of feminine and masculine words. In our CoT prompts, we require the LLM to explicitly indicate whether each word in the word list is a feminine or masculine before making the final predictions. With counting and handling the meaning of words, this benchmark has characteristics of both arithmetic reasoning and symbolic reasoning. Experimental results in English show that without step-by-step prediction, most LLMs make socially biased predictions, despite the task being as simple as counting words. Interestingly, CoT prompting reduces this unconscious social bias in LLMs and encourages fair predictions.

The Gaps between Pre-train and Downstream Settings in Bias Evaluation and Debiasing

The output tendencies of Pre-trained Language Models (PLM) vary markedly before and after Fine-Tuning (FT) due to the updates to the model parameters. These divergences in output tendencies result in a gap in the social biases of PLMs. For example, there exits a low correlation between intrinsic bias scores of a PLM and its extrinsic bias scores under FT-based debiasing methods. Additionally, applying FT-based debiasing methods to a PLM leads to a decline in performance in downstream tasks. On the other hand, PLMs trained on large datasets can learn without parameter updates via In-Context Learning (ICL) using prompts. ICL induces smaller changes to PLMs compared to FT-based debiasing methods. Therefore, we hypothesize that the gap observed in pre-trained and FT models does not hold true for debiasing methods that use ICL. In this study, we demonstrate that ICL-based debiasing methods show a higher correlation between intrinsic and extrinsic bias scores compared to FT-based methods. Moreover, the performance degradation due to debiasing is also lower in the ICL case compared to that in the FT case.

SAIE Framework: Support Alone Isn't Enough -- Advancing LLM Training with Adversarial Remarks

Large Language Models (LLMs) can justify or criticize their predictions through discussion with other models or humans, thereby enhancing their intrinsic understanding of instances. While proactive discussions enhance performance, this approach is currently limited to the inference phase. In this context, we posit a hypothesis: learning interactive discussions during training can improve understanding for the instances in the training step and proficiency in logical/critical thinking ability and verbalized expression of the model in the inference step. Our proposed SAIE training method involves both supportive and adversarial discussions between the learner and partner models. The learner model receives a remark from the partner through the discussion, and the parameters of the learner model are then updated based on this remark. That is, the teacher signal dynamically adjusts in response to the evolving model output throughout the training step. By bolstering the capacity for discussion and comprehension of instances, our experiments across datasets, including GSM8K, CommonsenseQA, and MMLU, reveal that models fine-tuned with our method consistently surpass those trained with standard fine-tuning techniques. Moreover, our approach demonstrates superior performance in multi-agent inference scenarios, boosting the models' reasoning abilities at the inference step.

How You Prompt Matters! Even Task-Oriented Constraints in Instructions Affect LLM-Generated Text Detection

Against the misuse (e.g., plagiarism or spreading misinformation) of Large Language Models (LLMs), many recent works have presented LLM-generated-text detectors with promising detection performance. Spotlighting a situation where users instruct LLMs to generate texts (e.g., essay writing), there are various ways to write the instruction (e.g., what task-oriented constraint to include). In this paper, we discover that even a task-oriented constraint in instruction can cause the inconsistent performance of current detectors to the generated texts. Specifically, we focus on student essay writing as a realistic domain and manually create the task-oriented constraint for each factor on essay quality by Ke and Ng (2019). Our experiment shows that the detection performance variance of the current detector on texts generated by instruction with each task-oriented constraint is up to 20 times larger than the variance caused by generating texts multiple times and paraphrasing the instruction. Our finding calls for further research on developing robust detectors that can detect such distributional shifts caused by a task-oriented constraint in the instruction.

Controlled Generation with Prompt Insertion for Natural Language Explanations in Grammatical Error Correction

In Grammatical Error Correction (GEC), it is crucial to ensure the user's comprehension of a reason for correction. Existing studies present tokens, examples, and hints as to the basis for correction but do not directly explain the reasons for corrections. Although methods that use Large Language Models (LLMs) to provide direct explanations in natural language have been proposed for various tasks, no such method exists for GEC. Generating explanations for GEC corrections involves aligning input and output tokens, identifying correction points, and presenting corresponding explanations consistently. However, it is not straightforward to specify a complex format to generate explanations, because explicit control of generation is difficult with prompts. This study introduces a method called controlled generation with Prompt Insertion (PI) so that LLMs can explain the reasons for corrections in natural language. In PI, LLMs first correct the input text, and then we automatically extract the correction points based on the rules. The extracted correction points are sequentially inserted into the LLM's explanation output as prompts, guiding the LLMs to generate explanations for the correction points. We also create an Explainable GEC (XGEC) dataset of correction reasons by annotating NUCLE, CoNLL2013, and CoNLL2014. Although generations from GPT-3 and ChatGPT using original prompts miss some correction points, the generation control using PI can explicitly guide to describe explanations for all correction points, contributing to improved performance in generating correction reasons.