The ability to recognize patterns from examples and apply them to new ones is a primal ability for general intelligence, and is widely studied by psychology and AI researchers. Many benchmarks have been proposed to measure such ability for Large Language Models (LLMs); however, they focus on few-shot (usually <10) setting and lack evaluation for aggregating many pieces of information from long contexts. On the other hand, the ever-growing context length of LLMs have brought forth the novel paradigm of many-shot In-Context Learning (ICL), which addresses new tasks with hundreds to thousands of examples without expensive and inefficient fine-tuning. However, many-shot evaluations often focus on classification, and popular long-context LLM tasks such as Needle-In-A-Haystack (NIAH) seldom require complicated intelligence for integrating many pieces of information. To fix the issues from both worlds, we propose MIR-Bench, the first many-shot in-context reasoning benchmark for pattern recognition that asks LLM to predict output via input-output examples from underlying functions with diverse data format. Based on MIR-Bench, we study many novel problems for many-shot in-context reasoning, and acquired many insightful findings including scaling effect, robustness, inductive vs. transductive reasoning, retrieval Augmented Generation (RAG), coding for inductive reasoning, cross-domain generalizability, etc.
We introduce DebateBench, a novel dataset consisting of an extensive collection of transcripts and metadata from some of the world's most prestigious competitive debates. The dataset consists of British Parliamentary debates from prestigious debating tournaments on diverse topics, annotated with detailed speech-level scores and house rankings sourced from official adjudication data. We curate 256 speeches across 32 debates with each debate being over 1 hour long with each input being an average of 32,000 tokens. Designed to capture long-context, large-scale reasoning tasks, DebateBench provides a benchmark for evaluating modern large language models (LLMs) on their ability to engage in argumentation, deliberation, and alignment with human experts. To do well on DebateBench, the LLMs must perform in-context learning to understand the rules and evaluation criteria of the debates, then analyze 8 seven minute long speeches and reason about the arguments presented by all speakers to give the final results. Our preliminary evaluation using GPT o1, GPT-4o, and Claude Haiku, shows that LLMs struggle to perform well on DebateBench, highlighting the need to develop more sophisticated techniques for improving their performance.
Although Large Language Models (LLMs) excel in reasoning and generation for language tasks, they are not specifically designed for multimodal challenges. Training Multimodal Large Language Models (MLLMs), however, is resource-intensive and constrained by various training limitations. In this paper, we propose the Modular-based Visual Contrastive Decoding (MVCD) framework to move this obstacle. Our framework leverages LLMs' In-Context Learning (ICL) capability and the proposed visual contrastive-example decoding (CED), specifically tailored for this framework, without requiring any additional training. By converting visual signals into text and focusing on contrastive output distributions during decoding, we can highlight the new information introduced by contextual examples, explore their connections, and avoid over-reliance on prior encoded knowledge. MVCD enhances LLMs' visual perception to make it see and reason over the input visuals. To demonstrate MVCD's effectiveness, we conduct experiments with four LLMs across five question answering datasets. Our results not only show consistent improvement in model accuracy but well explain the effective components inside our decoding strategy. Our code will be available at https://github.com/Pbhgit/MVCD.
We introduce a simple method for probabilistic predictions on tabular data based on Large Language Models (LLMs) called JoLT (Joint LLM Process for Tabular data). JoLT uses the in-context learning capabilities of LLMs to define joint distributions over tabular data conditioned on user-specified side information about the problem, exploiting the vast repository of latent problem-relevant knowledge encoded in LLMs. JoLT defines joint distributions for multiple target variables with potentially heterogeneous data types without any data conversion, data preprocessing, special handling of missing data, or model training, making it accessible and efficient for practitioners. Our experiments show that JoLT outperforms competitive methods on low-shot single-target and multi-target tabular classification and regression tasks. Furthermore, we show that JoLT can automatically handle missing data and perform data imputation by leveraging textual side information. We argue that due to its simplicity and generality, JoLT is an effective approach for a wide variety of real prediction problems.
In-context machine translation (MT) with large language models (LLMs) is a promising approach for low-resource MT, as it can readily take advantage of linguistic resources such as grammar books and dictionaries. Such resources are usually selectively integrated into the prompt so that LLMs can directly perform translation without any specific training, via their in-context learning capability (ICL). However, the relative importance of each type of resource e.g., dictionary, grammar book, and retrieved parallel examples, is not entirely clear. To address this gap, this study systematically investigates how each resource and its quality affects the translation performance, with the Manchu language as our case study. To remove any prior knowledge of Manchu encoded in the LLM parameters and single out the effect of ICL, we also experiment with an encrypted version of Manchu texts. Our results indicate that high-quality dictionaries and good parallel examples are very helpful, while grammars hardly help. In a follow-up study, we showcase a promising application of in-context MT: parallel data augmentation as a way to bootstrap the conventional MT model. When monolingual data abound, generating synthetic parallel data through in-context MT offers a pathway to mitigate data scarcity and build effective and efficient low-resource neural MT systems.
Large Language Models (LLMs) serve not only as chatbots but as key components in agent systems, where their common-sense knowledge significantly impacts performance as language-based planners for situated or embodied action. We assess LLMs' incremental learning (based on feedback from the environment), and controlled in-context learning abilities using a text-based environment. We introduce challenging yet interesting set of experiments to test i) how agents can incrementally solve tasks related to every day objects in typical rooms in a house where each of them are discovered by interacting within the environment, ii) controlled in-context learning abilities and efficiency of agents by providing short info about locations of objects and rooms to check how faster the task can be solved, and finally iii) using synthetic pseudo-English words to gauge how well LLMs are at inferring meaning of unknown words from environmental feedback. Results show that larger commercial models have a substantial gap in performance compared to open-weight but almost all models struggle with the synthetic words experiments.
Alignment tuning is crucial for ensuring large language models (LLMs) behave ethically and helpfully. Current alignment approaches require high-quality annotations and significant training resources. This paper proposes a low-cost, tuning-free method using in-context learning (ICL) to enhance LLM alignment. Through an analysis of high-quality ICL demos, we identified style as a key factor influencing LLM alignment capabilities and explicitly restyled ICL exemplars based on this stylistic framework. Additionally, we combined the restyled demos to achieve a balance between the two conflicting aspects of LLM alignment--factuality and safety. We packaged the restyled examples as prompts to trigger few-shot learning, improving LLM alignment. Compared to the best baseline approach, with an average score of 5.00 as the maximum, our method achieves a maximum 0.10 increase on the Alpaca task (from 4.50 to 4.60), a 0.22 enhancement on the Just-eval benchmark (from 4.34 to 4.56), and a maximum improvement of 0.32 (from 3.53 to 3.85) on the MT-Bench dataset. We release the code and data at https://github.com/AnonymousCode-ComputerScience/RIDE.
Planning and acting to solve `real' tasks using large language models (LLMs) in interactive environments has become a new frontier for AI methods. While recent advances allowed LLMs to interact with online tools, solve robotics tasks and many more, long range reasoning tasks remain a problem for LLMs. Existing methods to address this issue are very resource intensive and require additional data or human crafted rules, instead, we propose a simple method based on few-shot in-context learning alone to enhance `chain-of-thought' with state-tracking for planning and acting with LLMs. We show that our method establishes the new state-of-the-art on Alfworld for in-context learning methods (+14\% over the previous best few-shot in-context learning method) and performs on par with methods that use additional training data and additional tools such as code-execution. We also demonstrate that our enhanced `chain-of-states' allows the agent to both solve longer horizon problems and to be more efficient in number of steps required to solve a task. We show that our method works across a variety of LLMs for both API-based and open source ones. Finally, we also conduct ablation studies and show that `chain-of-thoughts' helps state-tracking accuracy, while a json-structure harms overall performance. We open-source our code and annotations at https://github.com/ai-nikolai/StateAct.
Length generalization, the ability to solve problems longer than those seen during training, remains a critical challenge for large language models (LLMs). Previous work modifies positional encodings (PEs) and data formats to improve length generalization on specific symbolic tasks such as addition and sorting. However, these approaches are fundamentally limited to special tasks, often degrading general language performance. Furthermore, they are typically evaluated on small transformers trained from scratch on single tasks and can cause performance drop when applied during post-training stage of practical LLMs with general capabilities. Hu et al., (2024) proposed Rule-Following Fine-Tuning (RFFT) to improve length generalization in the post-training stage of LLMs. Despite its compatibility with practical models and strong performance, RFFT is proposed for single tasks too, requiring re-training for each individual task with extensive examples. In this paper, we study length generalization in multi-task settings and propose Meta Rule-Following Fine-Tuning (Meta-RFFT), the first framework enabling robust cross-task length generalization. As our first contribution, we construct a large length generalization dataset containing 86 tasks spanning code execution, number processing, symbolic and logical reasoning tasks, beyond the common addition or multiplication tasks. Secondly, we show that cross-task length generalization is possible with Meta-RFFT. After training on a large number of tasks and instances, the models achieve remarkable length generalization ability on unseen tasks with minimal fine-tuning or one-shot prompting. For example, after fine-tuning on 1 to 5 digit addition, our 32B model achieves 95% accuracy on 30 digit addition, significantly outperforming the state-of-the-art reasoning models (DeepSeek-R1-671B: 72%), despite never seeing this task during RF-pretraining.
Multilingual large language models (MLLMs) are able to leverage in-context learning (ICL) to achieve high performance by leveraging cross-lingual knowledge transfer without parameter updates. However, their effectiveness is highly sensitive to example selection, particularly in multilingual settings. Based on the findings of existing work, three key factors influence multilingual ICL: (1) semantic similarity, (2) linguistic alignment, and (3) language-specific performance. However, existing approaches address these factors independently, without explicitly disentangling their combined impact, leaving optimal example selection underexplored. To address this gap, we propose balanced multi-factor ICL (\textbf{BMF-ICL}), a method that quantifies and optimally balances these factors for improved example selection. Experiments on mCSQA and TYDI across four MLLMs demonstrate that BMF-ICL outperforms existing methods. Further analysis highlights the importance of incorporating all three factors and the importance of selecting examples from multiple languages.
Text-to-SQL aims to convert natural language questions into executable SQL queries. While previous approaches, such as skeleton-masked selection, have demonstrated strong performance by retrieving similar training examples to guide large language models (LLMs), they struggle in real-world scenarios where such examples are unavailable. To overcome this limitation, we propose Self-Augmentation in-context learning with Fine-grained Example selection for Text-to-SQL (SAFE-SQL), a novel framework that improves SQL generation by generating and filtering self-augmented examples. SAFE-SQL first prompts an LLM to generate multiple Text-to-SQL examples relevant to the test input. Then SAFE-SQL filters these examples through three relevance assessments, constructing high-quality in-context learning examples. Using self-generated examples, SAFE-SQL surpasses the previous zero-shot, and few-shot Text-to-SQL frameworks, achieving higher execution accuracy. Notably, our approach provides additional performance gains in extra hard and unseen scenarios, where conventional methods often fail.
While multilingual large language models generally perform adequately, and sometimes even rival English performance on high-resource languages (HRLs), they often significantly underperform on low-resource languages (LRLs). Among several prompting strategies aiming at bridging the gap, multilingual in-context learning (ICL) has been particularly effective when demonstration in target languages is unavailable. However, there lacks a systematic understanding when and why it works well. In this work, we systematically analyze multilingual ICL, using demonstrations in HRLs to enhance cross-lingual transfer. We show that demonstrations in mixed HRLs consistently outperform English-only ones across the board, particularly for tasks written in LRLs. Surprisingly, our ablation study show that the presence of irrelevant non-English sentences in the prompt yields measurable gains, suggesting the effectiveness of multilingual exposure itself. Our results highlight the potential of strategically leveraging multilingual resources to bridge the performance gap for underrepresented languages.
Modern large language models (LLMs) employ various forms of logical inference, both implicitly and explicitly, when addressing reasoning tasks. Understanding how to optimally leverage these inference paradigms is critical for advancing LLMs' reasoning capabilities. This paper adopts an exploratory approach by introducing a controlled evaluation environment for analogical reasoning -- a fundamental cognitive task -- that is systematically parameterized across three dimensions: modality (textual, visual, symbolic), difficulty (easy, medium, hard), and task format (multiple-choice or free-text generation). We analyze the comparative dynamics of inductive, abductive, and deductive inference pipelines across these dimensions, and demonstrate that our findings generalize to broader in-context learning tasks. Additionally, we investigate advanced paradigms such as hypothesis selection, verification, and refinement, revealing their potential to scale up logical inference in LLM reasoning. This exploratory study provides a foundation for future research in enhancing LLM reasoning through systematic logical inference strategies.
The conversion of natural language into SQL language for querying databases (Text-to-SQL) has broad application prospects and has attracted widespread attention. At present, the mainstream Text-to-SQL methods are mainly divided into in-context learning (ICL) based methods and supervised fine-tuning (SFT) based methods. ICL-based methods can achieve relatively good results thanks to the use of the most advanced closed-source models. However, in real-world application scenarios, factors such as data privacy, SQL generation efficiency and cost need to be considered. SFT-based methods have certain advantages. At present, methods based on fine-tuning of open source models lack easy-to-implement and effective (cost-effective) baseline methods. We propose a pipeline-based method using open source model fine-tuning, referred to as BASE-SQL, which includes four components: Schema Linking, Candidate SQL Generate, SQL Revision and SQL Merge Revision. Experimental results show that BASE-SQL uses the open source model Qwen2.5-Coder-32B-Instruct, and achieves an accuracy of 67.47% on the BIRD development set and 88.9% on the Spider test set, which is significantly better than other methods using open source models, and even exceeds several methods using the GPT-4o closed-source model. At the same time, BASE-SQL is easy to implement and highly efficient (on average, only five calls to the large language model are required to generate SQL once). The code will be open sourced at https://github.com/CycloneBoy/base_sql.
By simply incorporating demonstrations into the context, in-context learning (ICL) enables large language models (LLMs) to yield awesome performance on many tasks. In this paper, we focus on passage-level long-context ICL for generation tasks and find that LLMs cannot learn the intrinsic relationships between the demonstration passage and the generation output. We conduct experiments with different LLMs on two typical generation tasks including single-document QA and distractor generation, demonstrating that even a completely meaningless demonstration passage with 1/4 length achieves much better performance than the original full passage. Analysis via attention score reveals that LLMs pay little attention to passages compared to other components in prompt and little attention flows from the passage to other parts of the demonstration, which further confirms our finding. Additionally, experiments on context compression indicate that compression approaches proven effective on other long-context tasks are not suitable for passage-level ICL, since simply using shorter meaningless demonstration passages has achieved competitive performance.
While transformer-based language models have driven the AI revolution thus far, their computational complexity has spurred growing interest in viable alternatives, such as structured state space sequence models (SSMs) and Selective SSMs. Among these, Mamba (S6) and its variant Mamba-2 have shown remarkable inference speed ups over transformers while achieving comparable or superior performance on complex language modeling tasks. However, despite these architectural innovations and empirical successes, the fundamental learning capabilities of Mamba remain poorly understood. In this paper, we address this gap by studying in-context learning (ICL) on Markov chains and uncovering a surprising phenomenon: unlike transformers, even a single-layer Mamba efficiently learns the in-context Laplacian smoothing estimator, which is both Bayes and minimax optimal, for all Markovian orders. To explain this, we theoretically characterize the representation capacity of Mamba and reveal the fundamental role of convolution in enabling it to represent the optimal Laplacian smoothing. These theoretical insights align strongly with empirical results and, to the best of our knowledge, represent the first formal connection between Mamba and optimal statistical estimators. Finally, we outline promising research directions inspired by these findings.
In this paper, we reexamine prompt engineering for large language models through the lens of automata theory. We argue that language models function as automata and, like all automata, should be programmed in the languages they accept, a unified collection of all natural and formal languages. Therefore, traditional software engineering practices--conditioned on the clear separation of programming languages and natural languages--must be rethought. We introduce the Ann Arbor Architecture, a conceptual framework for agent-oriented programming of language models, as a higher-level abstraction over raw token generation, and provide a new perspective on in-context learning. Based on this framework, we present the design of our agent platform Postline, and report on our initial experiments in agent training.
Memes present unique moderation challenges due to their subtle, multimodal interplay of images, text, and social context. Standard systems relying predominantly on explicit textual cues often overlook harmful content camouflaged by irony, symbolism, or cultural references. To address this gap, we introduce MemeSense, an adaptive in-context learning framework that fuses social commonsense reasoning with visually and semantically related reference examples. By encoding crucial task information into a learnable cognitive shift vector, MemeSense effectively balances lexical, visual, and ethical considerations, enabling precise yet context-aware meme intervention. Extensive evaluations on a curated set of implicitly harmful memes demonstrate that MemeSense substantially outperforms strong baselines, paving the way for safer online communities. Code and data available at: https://github.com/sayantan11995/MemeSense
This paper investigates approximation-theoretic aspects of the in-context learning capability of the transformers in representing a family of noisy linear dynamical systems. Our first theoretical result establishes an upper bound on the approximation error of multi-layer transformers with respect to an $L^2$-testing loss uniformly defined across tasks. This result demonstrates that transformers with logarithmic depth can achieve error bounds comparable with those of the least-squares estimator. In contrast, our second result establishes a non-diminishing lower bound on the approximation error for a class of single-layer linear transformers, which suggests a depth-separation phenomenon for transformers in the in-context learning of dynamical systems. Moreover, this second result uncovers a critical distinction in the approximation power of single-layer linear transformers when learning from IID versus non-IID data.
Grammatical error correction (GEC) aims to correct grammatical, spelling, and semantic errors in natural language text. With the growing of large language models (LLMs), direct text generation has gradually become the focus of the GEC methods, and few-shot in-context learning presents a cost-effective solution. However, selecting effective in-context examples remains challenging, as the similarity between input texts does not necessarily correspond to similar grammatical error patterns. In this paper, we propose a novel retrieval method based on natural language grammatical error explanations (GEE) to address this issue. Our method retrieves suitable few-shot demonstrations by matching the GEE of the test input with that of pre-constructed database samples, where explanations for erroneous samples are generated by LLMs. We conducted multilingual GEC few-shot experiments on both major open-source and closed-source LLMs. Experiments across five languages show that our method outperforms existing semantic and BM25-based retrieval techniques, without requiring additional training or language adaptation. This also suggests that matching error patterns is key to selecting examples.