Previous studies have shown that demonstrations can significantly help Large Language Models (LLMs ) perform better on the given tasks. However, this so-called In-Context Learning ( ICL ) ability is very sensitive to the presenting context, and often dozens of demonstrations are needed. In this work, we investigate if we can reduce the shot number while still maintaining a competitive performance. We present SeCoKD, a self-Knowledge Distillation ( KD ) training framework that aligns the student model with a heavily prompted variation, thereby increasing the utilization of a single demonstration. We experiment with the SeCoKD across three LLMs and six benchmarks focusing mainly on reasoning tasks. Results show that our method outperforms the base model and Supervised Fine-tuning ( SFT ), especially in zero-shot and one-shot settings by 30% and 10%, respectively. Moreover, SeCoKD brings little negative artifacts when evaluated on new tasks, which is more robust than Supervised Fine-tuning.
One way to address safety risks from large language models (LLMs) is to censor dangerous knowledge from their training data. While this removes the explicit information, implicit information can remain scattered across various training documents. Could an LLM infer the censored knowledge by piecing together these implicit hints? As a step towards answering this question, we study inductive out-of-context reasoning (OOCR), a type of generalization in which LLMs infer latent information from evidence distributed across training documents and apply it to downstream tasks without in-context learning. Using a suite of five tasks, we demonstrate that frontier LLMs can perform inductive OOCR. In one experiment we finetune an LLM on a corpus consisting only of distances between an unknown city and other known cities. Remarkably, without in-context examples or Chain of Thought, the LLM can verbalize that the unknown city is Paris and use this fact to answer downstream questions. Further experiments show that LLMs trained only on individual coin flip outcomes can verbalize whether the coin is biased, and those trained only on pairs $(x,f(x))$ can articulate a definition of $f$ and compute inverses. While OOCR succeeds in a range of cases, we also show that it is unreliable, particularly for smaller LLMs learning complex structures. Overall, the ability of LLMs to "connect the dots" without explicit in-context learning poses a potential obstacle to monitoring and controlling the knowledge acquired by LLMs.
Large Language Models (LLMs) have demonstrated impressive in-context learning (ICL) capabilities from few-shot demonstration exemplars. While recent learning-based demonstration selection methods have proven beneficial to ICL by choosing more useful exemplars, their underlying mechanisms are opaque, hindering efforts to address limitations such as high training costs and poor generalization across tasks. These methods generally assume the selection process captures similarities between the exemplar and the target instance, however, it remains unknown what kinds of similarities are captured and vital to performing ICL. To dive into this question, we analyze the working mechanisms of the learning-based demonstration selection methods and empirically identify two important factors related to similarity measurement: 1) The ability to integrate different levels of task-agnostic text similarities between the input of exemplars and test cases enhances generalization power across different tasks. 2) Incorporating task-specific labels when measuring the similarities significantly improves the performance on each specific task. We validate these two findings through extensive quantitative and qualitative analyses across ten datasets and various LLMs. Based on our findings, we introduce two effective yet simplified exemplar selection methods catering to task-agnostic and task-specific demands, eliminating the costly LLM inference overhead.
Language models often struggle with handling factual knowledge, exhibiting factual hallucination issue. This makes it vital to evaluate the models' ability to recall its parametric knowledge about facts. In this study, we introduce a knowledge probing benchmark, BELIEF(ICL), to evaluate the knowledge recall ability of both encoder- and decoder-based pre-trained language models (PLMs) from diverse perspectives. BELIEFs utilize a multi-prompt dataset to evaluate PLM's accuracy, consistency, and reliability in factual knowledge recall. To enable a more reliable evaluation with BELIEFs, we semi-automatically create MyriadLAMA, which has massively diverse prompts. We validate the effectiveness of BELIEFs in comprehensively evaluating PLM's knowledge recall ability on diverse PLMs, including recent large language models (LLMs). We then investigate key factors in memorizing and recalling facts in PLMs, such as model size, pretraining strategy and corpora, instruction-tuning process and in-context learning settings. Finally, we reveal the limitation of the prompt-based knowledge probing. The MyriadLAMA is publicized.
Large language models (LLMs) have demonstrated impressive capabilities, but still suffer from inconsistency issues (e.g. LLMs can react differently to disturbances like rephrasing or inconsequential order change). In addition to these inconsistencies, we also observe that LLMs, while capable of solving hard problems, can paradoxically fail at easier ones. To evaluate this hard-to-easy inconsistency, we develop the ConsisEval benchmark, where each entry comprises a pair of questions with a strict order of difficulty. Furthermore, we introduce the concept of consistency score to quantitatively measure this inconsistency and analyze the potential for improvement in consistency by relative consistency score. Based on comprehensive experiments across a variety of existing models, we find: (1) GPT-4 achieves the highest consistency score of 92.2\% but is still inconsistent to specific questions due to distraction by redundant information, misinterpretation of questions, etc.; (2) models with stronger capabilities typically exhibit higher consistency, but exceptions also exist; (3) hard data enhances consistency for both fine-tuning and in-context learning. Our data and code will be publicly available on GitHub.
In-context learning is a powerful capability of certain machine learning models that arguably underpins the success of today's frontier AI models. However, in-context learning is critically limited to settings where the in-context distribution of interest $p_{\theta}^{ICL}( x|\mathcal{D})$ can be straightforwardly expressed and/or parameterized by the model; for instance, language modeling relies on expressing the next-token distribution as a categorical distribution parameterized by the network's output logits. In this work, we present a more general form of in-context learning without such a limitation that we call \textit{in-context learning of energy functions}. The idea is to instead learn the unconstrained and arbitrary in-context energy function $E_{\theta}^{ICL}(x|\mathcal{D})$ corresponding to the in-context distribution $p_{\theta}^{ICL}(x|\mathcal{D})$. To do this, we use classic ideas from energy-based modeling. We provide preliminary evidence that our method empirically works on synthetic data. Interestingly, our work contributes (to the best of our knowledge) the first example of in-context learning where the input space and output space differ from one another, suggesting that in-context learning is a more-general capability than previously realized.
Recently, Large Language Models (LLMs) attained impressive performance in math and reasoning benchmarks. However, they still often struggle with logic problems and puzzles that are relatively easy for humans. To further investigate this, we introduce a new benchmark, SearchBench, containing 11 unique search problem types, each equipped with automated pipelines to generate an arbitrary number of instances and analyze the feasibility, correctness, and optimality of LLM-generated solutions. We show that even the most advanced LLMs fail to solve these problems end-to-end in text, e.g. GPT4 solves only 1.4%. SearchBench problems require considering multiple pathways to the solution as well as backtracking, posing a significant challenge to auto-regressive models. Instructing LLMs to generate code that solves the problem helps, but only slightly, e.g., GPT4's performance rises to 11.7%. In this work, we show that in-context learning with A* algorithm implementations enhances performance. The full potential of this promoting approach emerges when combined with our proposed Multi-Stage-Multi-Try method, which breaks down the algorithm implementation into two stages and verifies the first stage against unit tests, raising GPT-4's performance above 57%.
The widespread popularity of Large Language Models (LLMs), partly due to their unique ability to perform in-context learning, has also brought to light the importance of ethical and safety considerations when deploying these pre-trained models. In this work, we focus on investigating machine unlearning for LLMs motivated by data protection regulations. In contrast to the growing literature on fine-tuning methods to achieve unlearning, we focus on a comparatively lightweight alternative called soft prompting to realize the unlearning of a subset of training data. With losses designed to enforce forgetting as well as utility preservation, our framework \textbf{S}oft \textbf{P}rompting for \textbf{U}n\textbf{l}earning (SPUL) learns prompt tokens that can be appended to an arbitrary query to induce unlearning of specific examples at inference time without updating LLM parameters. We conduct a rigorous evaluation of the proposed method and our results indicate that SPUL can significantly improve the trade-off between utility and forgetting in the context of text classification and question answering with LLMs. We further validate our method using multiple LLMs to highlight the scalability of our framework and provide detailed insights into the choice of hyperparameters and the influence of the size of unlearning data. Our implementation is available at \url{https://github.com/karuna-bhaila/llm_unlearning}.
Recent developments show that Large Language Models (LLMs) produce state-of-the-art performance on natural language (NL) to code generation for resource-rich general-purpose languages like C++, Java, and Python. However, their practical usage for structured domain-specific languages (DSLs) such as YAML, JSON is limited due to domain-specific schema, grammar, and customizations generally unseen by LLMs during pre-training. Efforts have been made to mitigate this challenge via in-context learning through relevant examples or by fine-tuning. However, it suffers from problems, such as limited DSL samples and prompt sensitivity but enterprises maintain good documentation of the DSLs. Therefore, we propose DocCGen, a framework that can leverage such rich knowledge by breaking the NL-to-Code generation task for structured code languages into a two-step process. First, it detects the correct libraries using the library documentation that best matches the NL query. Then, it utilizes schema rules extracted from the documentation of these libraries to constrain the decoding. We evaluate our framework for two complex structured languages, Ansible YAML and Bash command, consisting of two settings: Out-of-domain (OOD) and In-domain (ID). Our extensive experiments show that DocCGen consistently improves different-sized language models across all six evaluation metrics, reducing syntactic and semantic errors in structured code. We plan to open-source the datasets and code to motivate research in constrained code generation.
Large Language Models (LLMs), already shown to ace various text comprehension tasks, have also remarkably been shown to tackle table comprehension tasks without specific training. Building on earlier studies of LLMs for tabular tasks, we probe how in-context learning (ICL), model scale, instruction tuning, and domain bias affect Tabular QA (TQA) robustness by testing LLMs, under diverse augmentations and perturbations, on diverse domains: Wikipedia-based $\textbf{WTQ}$, financial $\textbf{TAT-QA}$, and scientific $\textbf{SCITAB}$. Although instruction tuning and larger, newer LLMs deliver stronger, more robust TQA performance, data contamination and reliability issues, especially on $\textbf{WTQ}$, remain unresolved. Through an in-depth attention analysis, we reveal a strong correlation between perturbation-induced shifts in attention dispersion and the drops in performance, with sensitivity peaking in the model's middle layers. We highlight the need for improved interpretable methodologies to develop more reliable LLMs for table comprehension.
In scenarios where language models must incorporate new information efficiently without extensive retraining, traditional fine-tuning methods are prone to overfitting, degraded generalization, and unnatural language generation. To address these limitations, we introduce Consistent In-Context Editing (ICE), a novel approach leveraging the model's in-context learning capability to optimize toward a contextual distribution rather than a one-hot target. ICE introduces a simple yet effective optimization framework for the model to internalize new knowledge by aligning its output distributions with and without additional context. This method enhances the robustness and effectiveness of gradient-based tuning methods, preventing overfitting and preserving the model's integrity. We analyze ICE across four critical aspects of knowledge editing: accuracy, locality, generalization, and linguistic quality, demonstrating its advantages. Experimental results confirm the effectiveness of ICE and demonstrate its potential for continual editing, ensuring that the integrity of the model is preserved while updating information.
Misinformation is prevalent in various fields such as education, politics, health, etc., causing significant harm to society. However, current methods for cross-domain misinformation detection rely on time and resources consuming fine-tuning and complex model structures. With the outstanding performance of LLMs, many studies have employed them for misinformation detection. Unfortunately, they focus on in-domain tasks and do not incorporate significant sentiment and emotion features (which we jointly call affect). In this paper, we propose RAEmoLLM, the first retrieval augmented (RAG) LLMs framework to address cross-domain misinformation detection using in-context learning based on affective information. It accomplishes this by applying an emotion-aware LLM to construct a retrieval database of affective embeddings. This database is used by our retrieval module to obtain source-domain samples, which are subsequently used for the inference module's in-context few-shot learning to detect target domain misinformation. We evaluate our framework on three misinformation benchmarks. Results show that RAEmoLLM achieves significant improvements compared to the zero-shot method on three datasets, with the highest increases of 20.69%, 23.94%, and 39.11% respectively. This work will be released on https://github.com/lzw108/RAEmoLLM.
Effective organization of in-context learning (ICL) demonstrations is key to improving the quality of large language model (LLM) responses. To create better sample-label pairs that instruct LLM understanding, we introduce logit separability, a criterion to assess the clarity of both samples and class-related words at the logit level. This facilitates the optimization of sample and label selection, enhancing the precision of information provided in ICL demonstrations. Additionally, we find that incorporating multiple class-related words for each sample, rather than relying on a single class name, improves performance by offering a broader range of label information. Building on these insights, we propose LICL, a logit separability-based method that jointly organizes samples and integrates multiple class-related words into each sample-label pair. Evaluations across seven classification datasets show that this approach significantly improves ICL performance by providing clearer instructions and richer label information.
As Large Language Models (LLMs) are increasingly being employed in real-world applications in critical domains such as healthcare, it is important to ensure that the Chain-of-Thought (CoT) reasoning generated by these models faithfully captures their underlying behavior. While LLMs are known to generate CoT reasoning that is appealing to humans, prior studies have shown that these explanations do not accurately reflect the actual behavior of the underlying LLMs. In this work, we explore the promise of three broad approaches commonly employed to steer the behavior of LLMs to enhance the faithfulness of the CoT reasoning generated by LLMs: in-context learning, fine-tuning, and activation editing. Specifically, we introduce novel strategies for in-context learning, fine-tuning, and activation editing aimed at improving the faithfulness of the CoT reasoning. We then carry out extensive empirical analyses with multiple benchmark datasets to explore the promise of these strategies. Our analyses indicate that these strategies offer limited success in improving the faithfulness of the CoT reasoning, with only slight performance enhancements in controlled scenarios. Activation editing demonstrated minimal success, while fine-tuning and in-context learning achieved marginal improvements that failed to generalize across diverse reasoning and truthful question-answering benchmarks. In summary, our work underscores the inherent difficulty in eliciting faithful CoT reasoning from LLMs, suggesting that the current array of approaches may not be sufficient to address this complex challenge.
Existing in-context learning (ICL) methods for relation extraction (RE) often prioritize language similarity over structural similarity, which can lead to overlooking entity relationships. To address this, we propose an AMR-enhanced retrieval-based ICL method for RE. Our model retrieves in-context examples based on semantic structure similarity between task inputs and training samples. Evaluations on four standard English RE datasets show that our model outperforms baselines in the unsupervised setting across all datasets. In the supervised setting, it achieves state-of-the-art results on three datasets and competitive results on the fourth.
To seek reliable information sources for news events, we introduce a novel task of expert recommendation, which aims to identify trustworthy sources based on their previously quoted statements. To achieve this, we built a novel dataset, called NewsQuote, consisting of 23,571 quote-speaker pairs sourced from a collection of news articles. We formulate the recommendation task as the retrieval of experts based on their likelihood of being associated with a given query. We also propose a multi-layer ranking framework employing Large Language Models to improve the recommendation performance. Our results show that employing an in-context learning based LLM ranker and a multi-layer ranking-based filter significantly improve both the predictive quality and behavioural quality of the recommender system.
We introduce Meta-Reasoning Prompting (MRP), a novel and efficient system prompting method for large language models (LLMs) inspired by human meta-reasoning. Traditional in-context learning-based reasoning techniques, such as Tree-of-Thoughts, show promise but lack consistent state-of-the-art performance across diverse tasks due to their specialized nature. MRP addresses this limitation by guiding LLMs to dynamically select and apply different reasoning methods based on the specific requirements of each task, optimizing both performance and computational efficiency. With MRP, LLM reasoning operates in two phases. Initially, the LLM identifies the most appropriate reasoning method using task input cues and objective descriptions of available methods. Subsequently, it applies the chosen method to complete the task. This dynamic strategy mirrors human meta-reasoning, allowing the model to excel in a wide range of problem domains. We evaluate the effectiveness of MRP through comprehensive benchmarks. The results demonstrate that MRP achieves or approaches state-of-the-art performance across diverse tasks. MRP represents a significant advancement in enabling LLMs to identify cognitive challenges across problems and leverage benefits across different reasoning approaches, enhancing their ability to handle diverse and complex problem domains efficiently. Every LLM deserves a Meta-Reasoning Prompting to unlock its full potential and ensure adaptability in an ever-evolving landscape of challenges and applications.
Utilizing Large Language Models (LLMs) as evaluators to assess the performance of LLMs has garnered attention. However, this kind of evaluation approach is affected by potential biases within LLMs, raising concerns about the accuracy and reliability of the evaluation results of LLMs. To address this problem, we propose and study two many-shot In-Context Learning (ICL) prompt templates to help LLM evaluators mitigate potential biases: Many-Shot with Reference (MSwR) and Many-Shot without Reference (MSoR). Specifically, the former utilizes in-context examples with model-generated evaluation rationales as references, while the latter does not include these references. Using these prompt designs, we investigate the impact of increasing the number of in-context examples on the consistency and quality of the evaluation results. Experimental results show that advanced LLMs, such as GPT-4o, perform better in the many-shot regime than in the zero-shot and few-shot regimes. Furthermore, when using GPT-4o as an evaluator in the many-shot regime, adopting MSwR as the prompt template performs better than MSoR.
The reasoning abilities of Large Language Models (LLMs) are becoming a central focus of study in NLP. In this paper, we consider the case of syllogistic reasoning, an area of deductive reasoning studied extensively in logic and cognitive psychology. Previous research has shown that pre-trained LLMs exhibit reasoning biases, such as $\textit{content effects}$, avoid answering that $\textit{no conclusion follows}$, display human-like difficulties, and struggle with multi-step reasoning. We contribute to this research line by systematically investigating the effects of chain-of-thought reasoning, in-context learning (ICL), and supervised fine-tuning (SFT) on syllogistic reasoning, considering syllogisms with conclusions that support or violate world knowledge, as well as ones with multiple premises. Crucially, we go beyond the standard focus on accuracy, with an in-depth analysis of the conclusions generated by the models. Our results suggest that the behavior of pre-trained LLMs can be explained by heuristics studied in cognitive science and that both ICL and SFT improve model performance on valid inferences, although only the latter mitigates most reasoning biases without harming model consistency.
We present a method for rewriting an input sentence to match specific values of nontrivial linguistic features, such as dependency depth. In contrast to earlier work, our method uses in-context learning rather than finetuning, making it applicable in use cases where data is sparse. We show that our model performs accurate rewrites and matches the state of the art on rewriting sentences to a specified school grade level.