Scientific databases aggregate vast amounts of quantitative data alongside descriptive text. In biochemistry, molecule screening assays evaluate the functional responses of candidate molecules against disease targets. Unstructured text that describes the biological mechanisms through which these targets operate, experimental screening protocols, and other attributes of assays offer rich information for new drug discovery campaigns but has been untapped because of that unstructured format. We present Assay2Mol, a large language model-based workflow that can capitalize on the vast existing biochemical screening assays for early-stage drug discovery. Assay2Mol retrieves existing assay records involving targets similar to the new target and generates candidate molecules using in-context learning with the retrieved assay screening data. Assay2Mol outperforms recent machine learning approaches that generate candidate ligand molecules for target protein structures, while also promoting more synthesizable molecule generation.
Recent advancements in medical image analysis have led to the development of highly specialized models tailored to specific clinical tasks. These models have demonstrated exceptional performance and remain a crucial research direction. Yet, their applicability is limited to predefined tasks, requiring expertise and extensive resources for development and adaptation. In contrast, generalist models offer a different form of utility: allowing medical practitioners to define tasks on the fly without the need for task-specific model development. In this work, we explore how to train generalist models for the domain of retinal optical coherence tomography using visual in-context learning (VICL), i.e., training models to generalize across tasks based on a few examples provided at inference time. To facilitate rigorous assessment, we propose a broad evaluation protocol tailored to VICL in OCT. We extensively evaluate a state-of-the-art medical VICL approach on multiple retinal OCT datasets, establishing a first baseline to highlight the potential and current limitations of in-context learning for OCT. To foster further research and practical adoption, we openly release our code.
Vision-Language Models (VLMs) have rapidly advanced alongside Large Language Models (LLMs). This study evaluates the capabilities of prominent generative VLMs, such as GPT-4.1 and Gemini 2.5 Pro, accessed via APIs, for histopathology image classification tasks, including cell typing. Using diverse datasets from public and private sources, we apply zero-shot and one-shot prompting methods to assess VLM performance, comparing them against custom-trained Convolutional Neural Networks (CNNs). Our findings demonstrate that while one-shot prompting significantly improves VLM performance over zero-shot ($p \approx 1.005 \times 10^{-5}$ based on Kappa scores), these general-purpose VLMs currently underperform supervised CNNs on most tasks. This work underscores both the promise and limitations of applying current VLMs to specialized domains like pathology via in-context learning. All code and instructions for reproducing the study can be accessed from the repository https://www.github.com/a12dongithub/VLMCCE.
We propose a novel prompt design paradigm that challenges conventional wisdom in large language model (LLM) prompting. While conventional wisdom prioritizes well-crafted instructions and demonstrations for in-context learning (ICL), we show that pruning random demonstrations into seemingly incoherent "gibberish" can remarkably improve performance across diverse tasks. Notably, the "gibberish" always matches or surpasses state-of-the-art automatic prompt optimization techniques, achieving substantial gains regardless of LLM alignment. Nevertheless, discovering an effective pruning strategy is non-trivial, as existing attribution methods and prompt compression algorithms fail to deliver robust results, let alone human intuition. In terms of this, we propose a self-discover prompt optimization framework, PromptQuine, an evolutionary search framework that automatically searches for the pruning strategy by itself using only low-data regimes. Much like the emergent complexity in nature--such as symbiosis and self-organization--arising in response to resource constraints, our framework evolves and refines unconventional yet highly effective prompts by leveraging only the tokens present within the context. We demonstrate its effectiveness across classification, multi-choice question answering, generation and math reasoning tasks across LLMs, while achieving decent runtime efficiency. We hope our findings can guide mechanistic studies on in-context learning, and provide a call to action, to pave the way for more open-ended search algorithms for more effective LLM prompting.
Artificial Intelligence (AI) is rapidly embedded in critical decision-making systems, however their foundational ``black-box'' models require eXplainable AI (XAI) solutions to enhance transparency, which are mostly oriented to experts, making no sense to non-experts. Alarming evidence about AI's unprecedented human values risks brings forward the imperative need for transparent human-centered XAI solutions. In this work, we introduce a domain-, model-, explanation-agnostic, generalizable and reproducible framework that ensures both transparency and human-centered explanations tailored to the needs of both experts and non-experts. The framework leverages Large Language Models (LLMs) and employs in-context learning to convey domain- and explainability-relevant contextual knowledge into LLMs. Through its structured prompt and system setting, our framework encapsulates in one response explanations understandable by non-experts and technical information to experts, all grounded in domain and explainability principles. To demonstrate the effectiveness of our framework, we establish a ground-truth contextual ``thesaurus'' through a rigorous benchmarking with over 40 data, model, and XAI combinations for an explainable clustering analysis of a well-being scenario. Through a comprehensive quality and human-friendliness evaluation of our framework's explanations, we prove high content quality through strong correlations with ground-truth explanations (Spearman rank correlation=0.92) and improved interpretability and human-friendliness to non-experts through a user study (N=56). Our overall evaluation confirms trust in LLMs as HCXAI enablers, as our framework bridges the above Gaps by delivering (i) high-quality technical explanations aligned with foundational XAI methods and (ii) clear, efficient, and interpretable human-centered explanations for non-experts.
We investigate the use of randomly generated data for the sake of pre-training a model. We justify this approach theoretically from the perspective of algorithmic complexity, building on recent research that shows that sequence models can be trained to approximate Solomonoff induction. We derive similar, but complementary theoretical results. We show empirically that synthetically generated data can be used to pre-train a model before the data is seen. We replicate earlier results that models trained this way show zero-shot in-context learning across a variety of datasets, and that this performance improves with scale. We extend earlier results to real-world data, and show that finetuning a model after pre-training offers faster convergence and better generalization.
LLM-assisted hardware verification is gaining substantial attention due to its potential to significantly reduce the cost and effort of crafting effective testbenches. It also serves as a critical enabler for LLM-aided end-to-end hardware language design. However, existing current LLMs often struggle with Register Transfer Level (RTL) code generation, resulting in testbenches that exhibit functional errors in Hardware Description Languages (HDL) logic. Motivated by the strong performance of LLMs in Python code generation under inference-time sampling strategies, and their promising capabilities as judge agents, we propose PRO-V a fully program generation multi-agent system for robust RTL verification. Pro-V incorporates an efficient best-of-n iterative sampling strategy to enhance the correctness of generated testbenches. Moreover, it introduces an LLM-as-a-judge aid validation framework featuring an automated prompt generation pipeline. By converting rule-based static analysis from the compiler into natural language through in-context learning, this pipeline enables LLMs to assist the compiler in determining whether verification failures stem from errors in the RTL design or the testbench. PRO-V attains a verification accuracy of 87.17% on golden RTL implementations and 76.28% on RTL mutants. Our code is open-sourced at https://github.com/stable-lab/Pro-V.
Tabular in-context learning (ICL) has recently achieved state-of-the-art (SOTA) performance on several tabular prediction tasks. Previously restricted to classification problems on small tables, recent advances such as TabPFN and TabICL have extended its use to larger datasets. While being architecturally efficient and well-adapted to tabular data structures, current table-native ICL architectures, being trained exclusively on synthetic data, do not fully leverage the rich semantics and world knowledge contained in real-world tabular data. On another end of this spectrum, tabular ICL models based on pretrained large language models such as TabuLa-8B integrate deep semantic understanding and world knowledge but are only able to make use of a small amount of context due to inherent architectural limitations. With the aim to combine the best of both these worlds, we introduce ConTextTab, integrating semantic understanding and alignment into a table-native ICL framework. By employing specialized embeddings for different data modalities and by training on large-scale real-world tabular data, our model is competitive with SOTA across a broad set of benchmarks while setting a new standard on the semantically rich CARTE benchmark.
Foundation models are an emerging research direction in tabular deep learning. Notably, TabPFNv2 recently claimed superior performance over traditional GBDT-based methods on small-scale datasets using an in-context learning paradigm, which does not adapt model parameters to target datasets. However, the optimal finetuning approach for adapting tabular foundational models, and how this adaptation reshapes their internal mechanisms, remains underexplored. While prior works studied finetuning for earlier foundational models, inconsistent findings and TabPFNv2's unique architecture necessitate fresh investigation. To address these questions, we first systematically evaluate various finetuning strategies on diverse datasets. Our findings establish full finetuning as the most practical solution for TabPFNv2 in terms of time-efficiency and effectiveness. We then investigate how finetuning alters TabPFNv2's inner mechanisms, drawing an analogy to retrieval-augmented models. We reveal that the success of finetuning stems from the fact that after gradient-based adaptation, the dot products of the query-representations of test objects and the key-representations of in-context training objects more accurately reflect their target similarity. This improved similarity allows finetuned TabPFNv2 to better approximate target dependency by appropriately weighting relevant in-context samples, improving the retrieval-based prediction logic. From the practical perspective, we managed to finetune TabPFNv2 on datasets with up to 50K objects, observing performance improvements on almost all tasks. More precisely, on academic datasets with I.I.D. splits, finetuning allows TabPFNv2 to achieve state-of-the-art results, while on datasets with gradual temporal shifts and rich feature sets, TabPFNv2 is less stable and prior methods remain better.
The in-context learning paradigm with LLMs has been instrumental in advancing a wide range of natural language processing tasks. The selection of few-shot examples (exemplars / demonstration samples) is essential for constructing effective prompts under context-length budget constraints. In this paper, we formulate the exemplar selection task as a top-m best arms identification problem. A key challenge in this setup is the exponentially large number of arms that need to be evaluated to identify the m-best arms. We propose CASE (Challenger Arm Sampling for Exemplar selection), a novel sample-efficient selective exploration strategy that maintains a shortlist of "challenger" arms, which are current candidates for the top-m arms. In each iteration, only one of the arms from this shortlist or the current topm set is pulled, thereby reducing sample complexity and, consequently, the number of LLM evaluations. Furthermore, we model the scores of exemplar subsets (arms) using a parameterized linear scoring function, leading to stochastic linear bandits setting. CASE achieves remarkable efficiency gains of up to 7x speedup in runtime while requiring 7x fewer LLM calls (87% reduction) without sacrificing performance compared to state-of-the-art exemplar selection methods. We release our code and data at https://github.com/kiranpurohit/CASE
Large language models have transformed natural language processing, yet supervised fine-tuning (SFT) remains computationally intensive. This paper formally proves that capabilities acquired through SFT can be approximated by a base transformer model using inference-time techniques, specifically in-context learning (ICL), without altering model parameters, under idealized assumptions including unbounded computational resources and access to the fine-tuning dataset. We extend these results to practical scenarios with finite context lengths and partial dataset access. For text generation tasks with fixed output length $l$, datasets of size $\mathrm{O}\left( \frac{m V}{\varepsilon^2} \log \frac{m}{\delta} \right)$ or, with bounded context, $\mathrm{O}\left( \frac{l \log V}{\varepsilon^2} \log \frac{1}{\delta} \right)$ suffice to approximate fine-tuned behavior across $m$ contexts within error $\varepsilon$, where $V$ is the vocabulary size and $\delta$ is the failure probability. For linear classification, datasets of size $\mathrm{O}\left( \frac{d}{\varepsilon} \right)$ or, with fixed context, $\mathrm{O}\left( \frac{1}{\varepsilon^2} \log \frac{1}{\delta} \right)$ are sufficient, where $d$ is the input dimension. Grounded in the Turing completeness of transformers, these results provide a theoretical foundation for resource-efficient deployment of large language models, with practical techniques like retrieval-augmented generation bridging theory to real-world applications.
Causal effect estimation from observational data is fundamental across various applications. However, selecting an appropriate estimator from dozens of specialized methods demands substantial manual effort and domain expertise. We present CausalPFN, a single transformer that amortizes this workflow: trained once on a large library of simulated data-generating processes that satisfy ignorability, it infers causal effects for new observational datasets out-of-the-box. CausalPFN combines ideas from Bayesian causal inference with the large-scale training protocol of prior-fitted networks (PFNs), learning to map raw observations directly to causal effects without any task-specific adjustment. Our approach achieves superior average performance on heterogeneous and average treatment effect estimation benchmarks (IHDP, Lalonde, ACIC). Moreover, it shows competitive performance for real-world policy making on uplift modeling tasks. CausalPFN provides calibrated uncertainty estimates to support reliable decision-making based on Bayesian principles. This ready-to-use model does not require any further training or tuning and takes a step toward automated causal inference (https://github.com/vdblm/CausalPFN).
Large-scale neural language models (LMs) exhibit remarkable performance in in-context learning: the ability to learn and reason the input context on the fly without parameter update. This work studies in-context counterfactual reasoning in language models, that is, to predict the consequences of changes under hypothetical scenarios. We focus on studying a well-defined synthetic setup: a linear regression task that requires noise abduction, where accurate prediction is based on inferring and copying the contextual noise from factual observations. We show that language models are capable of counterfactual reasoning in this controlled setup and provide insights that counterfactual reasoning for a broad class of functions can be reduced to a transformation on in-context observations; we find self-attention, model depth, and data diversity in pre-training drive performance in Transformers. More interestingly, our findings extend beyond regression tasks and show that Transformers can perform noise abduction on sequential data, providing preliminary evidence on the potential for counterfactual story generation. Our code is available under https://github.com/moXmiller/counterfactual-reasoning.git .
In-Context Learning (ICL) empowers Large Language Models (LLMs) for rapid task adaptation without Fine-Tuning (FT), but its reliance on demonstration selection remains a critical challenge. While many-shot ICL shows promising performance through scaled demonstrations, the selection method for many-shot demonstrations remains limited to random selection in existing work. Since the conventional instance-level retrieval is not suitable for many-shot scenarios, we hypothesize that the data requirements for in-context learning and fine-tuning are analogous. To this end, we introduce a novel gradient matching approach that selects demonstrations by aligning fine-tuning gradients between the entire training set of the target task and the selected examples, so as to approach the learning effect on the entire training set within the selected examples. Through gradient matching on relatively small models, e.g., Qwen2.5-3B or Llama3-8B, our method consistently outperforms random selection on larger LLMs from 4-shot to 128-shot scenarios across 9 diverse datasets. For instance, it surpasses random selection by 4% on Qwen2.5-72B and Llama3-70B, and by around 2% on 5 closed-source LLMs. This work unlocks more reliable and effective many-shot ICL, paving the way for its broader application.
Neuro-symbolic approaches combining large language models (LLMs) with solvers excels in logical reasoning problems need long reasoning chains. In this paradigm, LLMs serve as translators, converting natural language reasoning problems into formal logic formulas. Then reliable symbolic solvers return correct solutions. Despite their success, we find that LLMs, as translators, struggle to handle lexical diversification, a common linguistic phenomenon, indicating that LLMs as logic translators are unreliable in real-world scenarios. Moreover, existing logical reasoning benchmarks lack lexical diversity, failing to challenge LLMs' ability to translate such text and thus obscuring this issue. In this work, we propose SCALe, a benchmark designed to address this significant gap through **logic-invariant lexical diversification**. By using LLMs to transform original benchmark datasets into lexically diversified but logically equivalent versions, we evaluate LLMs' ability to consistently map diverse expressions to uniform logical symbols on these new datasets. Experiments using SCALe further confirm that current LLMs exhibit deficiencies in this capability. Building directly on the deficiencies identified through our benchmark, we propose a new method, MenTaL, to address this limitation. This method guides LLMs to first construct a table unifying diverse expressions before performing translation. Applying MenTaL through in-context learning and supervised fine-tuning (SFT) significantly improves the performance of LLM translators on lexically diversified text. Our code is now available at https://github.com/wufeiwuwoshihua/LexicalDiver.
The Internet comprises of interconnected, independently managed Autonomous Systems (AS) that rely on the Border Gateway Protocol (BGP) for inter-domain routing. BGP anomalies--such as route leaks and hijacks--can divert traffic through unauthorized or inefficient paths, jeopardizing network reliability and security. Although existing rule-based and machine learning methods can detect these anomalies using structured metrics, they still require experts with in-depth BGP knowledge of, for example, AS relationships and historical incidents, to interpret events and propose remediation. In this paper, we introduce BEAR (BGP Event Analysis and Reporting), a novel framework that leverages large language models (LLMs) to automatically generate comprehensive reports explaining detected BGP anomaly events. BEAR employs a multi-step reasoning process that translates tabular BGP data into detailed textual narratives, enhancing interpretability and analytical precision. To address the limited availability of publicly documented BGP anomalies, we also present a synthetic data generation framework powered by LLMs. Evaluations on both real and synthetic datasets demonstrate that BEAR achieves 100% accuracy, outperforming Chain-of-Thought and in-context learning baselines. This work pioneers an automated approach for explaining BGP anomaly events, offering valuable operational insights for network management.
Large language models (LLMs) are trained on extensive datasets that encapsulate substantial world knowledge. However, their outputs often include confidently stated inaccuracies. Earlier works suggest that LLMs encode truthfulness as a distinct linear feature, termed the "truth direction", which can classify truthfulness reliably. We address several open questions about the truth direction: (i) whether LLMs universally exhibit consistent truth directions; (ii) whether sophisticated probing techniques are necessary to identify truth directions; and (iii) how the truth direction generalizes across diverse contexts. Our findings reveal that not all LLMs exhibit consistent truth directions, with stronger representations observed in more capable models, particularly in the context of logical negation. Additionally, we demonstrate that truthfulness probes trained on declarative atomic statements can generalize effectively to logical transformations, question-answering tasks, in-context learning, and external knowledge sources. Finally, we explore the practical application of truthfulness probes in selective question-answering, illustrating their potential to improve user trust in LLM outputs. These results advance our understanding of truth directions and provide new insights into the internal representations of LLM beliefs. Our code is public at https://github.com/colored-dye/truthfulness_probe_generalization
Forecasting over Temporal Knowledge Graphs (TKGs) which predicts future facts based on historical ones has received much attention. Recent studies have introduced Large Language Models (LLMs) for this task to enhance the models' generalization abilities. However, these models perform forecasting via simultaneously learning two kinds of entangled knowledge in the TKG: (1) general patterns, i.e., invariant temporal structures shared across different scenarios; and (2) scenario information, i.e., factual knowledge engaged in specific scenario, such as entities and relations. As a result, the learning processes of these two kinds of knowledge may interfere with each other, which potentially impact the generalization abilities of the models. To enhance the generalization ability of LLMs on this task, in this paper, we propose a General-to-Specific learning framework (G2S) that disentangles the learning processes of the above two kinds of knowledge. In the general learning stage, we mask the scenario information in different TKGs and convert it into anonymous temporal structures. After training on these structures, the model is able to capture the general patterns across different TKGs. In the specific learning stage, we inject the scenario information into the structures via either in-context learning or fine-tuning modes. Experimental results show that G2S effectively improves the generalization abilities of LLMs.
Detecting toxic content using language models is important but challenging. While large language models (LLMs) have demonstrated strong performance in understanding Chinese, recent studies show that simple character substitutions in toxic Chinese text can easily confuse the state-of-the-art (SOTA) LLMs. In this paper, we highlight the multimodal nature of Chinese language as a key challenge for deploying LLMs in toxic Chinese detection. First, we propose a taxonomy of 3 perturbation strategies and 8 specific approaches in toxic Chinese content. Then, we curate a dataset based on this taxonomy, and benchmark 9 SOTA LLMs (from both the US and China) to assess if they can detect perturbed toxic Chinese text. Additionally, we explore cost-effective enhancement solutions like in-context learning (ICL) and supervised fine-tuning (SFT). Our results reveal two important findings. (1) LLMs are less capable of detecting perturbed multimodal Chinese toxic contents. (2) ICL or SFT with a small number of perturbed examples may cause the LLMs "overcorrect'': misidentify many normal Chinese contents as toxic.
In-context learning (ICL) enables large language models (LLMs) to perform new tasks using only a few demonstrations. In Named Entity Recognition (NER), demonstrations are typically selected based on semantic similarity to the test instance, ignoring training labels and resulting in suboptimal performance. We introduce DEER, a new method that leverages training labels through token-level statistics to improve ICL performance. DEER first enhances example selection with a label-guided, token-based retriever that prioritizes tokens most informative for entity recognition. It then prompts the LLM to revisit error-prone tokens, which are also identified using label statistics, and make targeted corrections. Evaluated on five NER datasets using four different LLMs, DEER consistently outperforms existing ICL methods and approaches the performance of supervised fine-tuning. Further analysis shows its effectiveness on both seen and unseen entities and its robustness in low-resource settings.