Xunjian Yin, Sitao Cheng, Yuxi Xie, Xinyu Hu, Li Lin, Xinyi Wang, Liangming Pan, William Yang Wang, Xiaojun Wan. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Mingyu Jin, Weidi Luo, Sitao Cheng, Xinyi Wang, Wenyue Hua, Ruixiang Tang, William Yang Wang, Yongfeng Zhang. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

The rapid advancement of large language models (LLMs) has significantly enhanced the capabilities of agents across various tasks.However, existing agentic systems, whether based on fixed pipeline algorithms or pre-defined meta-learning frameworks, cannot search the whole agent design space due to the restriction of human-designed components, and thus might miss the more optimal agent design.In this paper, we introduce Gödel Agent, a selfevolving framework inspired by the Gödel machine, enabling agents to recursively improve themselves without relying on predefined routines or fixed optimization algorithms.Gödel Agent leverages LLMs to dynamically modify its own logic and behavior, guided solely by high-level objectives through prompting.Experimental results on multiple domains demonstrate that implementation of Gödel Agent can achieve continuous self-improvement, surpassing manually crafted agents in performance, efficiency, and generalizability.

The impressive capabilities of large language models (LLMs) have sparked debate over whether these models genuinely generalize to unseen tasks or predominantly rely on memorizing vast amounts of pretraining data. To explore this issue, we introduce an extended concept of memorization, distributional memorization, which measures the correlation between the LLM output probabilities and the pretraining data frequency. To effectively capture task-specific pretraining data frequency, we propose a novel task-gram language model, which is built by counting the co-occurrence of semantically related $n$-gram pairs from task inputs and outputs in the pretraining corpus. Using the Pythia models trained on the Pile dataset, we evaluate four distinct tasks: machine translation, factual question answering, world knowledge understanding, and math reasoning. Our findings reveal varying levels of memorization, with the strongest effect observed in factual question answering. Furthermore, while model performance improves across all tasks as LLM size increases, only factual question answering shows an increase in memorization, whereas machine translation and reasoning tasks exhibit greater generalization, producing more novel outputs. This study demonstrates that memorization plays a larger role in simpler, knowledge-intensive tasks, while generalization is the key for harder, reasoning-based tasks, providing a scalable method for analyzing large pretraining corpora in greater depth.

Pre-trained language models (LMs) are able to perform complex reasoning without explicit fine-tuning. To understand how pre-training with a next-token prediction objective contributes to the emergence of such reasoning capability, we propose that we can view an LM as deriving new conclusions by aggregating indirect reasoning paths seen at pre-training time. We found this perspective effective in two important cases of reasoning: logic reasoning with knowledge graphs (KGs) and chain-of-thought (CoT) reasoning. More specifically, we formalize the reasoning paths as random walk paths on the knowledge/reasoning graphs. Analyses of learned LM distributions suggest that a weighted sum of relevant random walk path probabilities is a reasonable way to explain how LMs reason. Experiments and analysis on multiple KG and CoT datasets reveal the effect of training on random walk paths and suggest that augmenting unlabeled random walk reasoning paths can improve real-world multi-step reasoning performance.

As the number of accepted papers at AI and ML conferences reaches into the thousands, it has become unclear how researchers access and read research publications. In this paper, we investigate the role of social media influencers in enhancing the visibility of machine learning research, particularly the citation counts of papers they share. We have compiled a comprehensive dataset of over 8,000 papers, spanning tweets from December 2018 to October 2023, alongside controls precisely matched by 9 key covariates. Our statistical and causal inference analysis reveals a significant increase in citations for papers endorsed by these influencers, with median citation counts 2-3 times higher than those of the control group. Additionally, the study delves into the geographic, gender, and institutional diversity of highlighted authors. Given these findings, we advocate for a responsible approach to curation, encouraging influencers to uphold the journalistic standard that includes showcasing diverse research topics, authors, and institutions.

Diffusion-based text-to-video (T2V) models have achieved significant success but continue to be hampered by the slow sampling speed of their iterative sampling processes. To address the challenge, consistency models have been proposed to facilitate fast inference, albeit at the cost of sample quality. In this work, we aim to break the quality bottleneck of a video consistency model (VCM) to achieve **both fast and high-quality video generation**. We introduce T2V-Turbo, which integrates feedback from a mixture of differentiable reward models into the consistency distillation (CD) process of a pre-trained T2V model. Notably, we directly optimize rewards associated with single-step generations that arise naturally from computing the CD loss, effectively bypassing the memory constraints imposed by backpropagating gradients through an iterative sampling process. Remarkably, the 4-step generations from our T2V-Turbo achieve the highest total score on VBench, even surpassing Gen-2 and Pika. We further conduct human evaluations to corroborate the results, validating that the 4-step generations from our T2V-Turbo are preferred over the 50-step DDIM samples from their teacher models, representing more than a tenfold acceleration while improving video generation quality.

Abstract While large language models (LLMs) have shown remarkable effectiveness in various NLP tasks, they are still prone to issues such as hallucination, unfaithful reasoning, and toxicity. A promising approach to rectify these flaws is correcting LLMs with feedback, where the LLM itself is prompted or guided with feedback to fix problems in its own output. Techniques leveraging automated feedback—either produced by the LLM itself (self-correction) or some external system—are of particular interest as they make LLM-based solutions more practical and deployable with minimal human intervention. This paper provides an exhaustive review of the recent advances in correcting LLMs with automated feedback, categorizing them into training-time, generation-time, and post-hoc approaches. We also identify potential challenges and future directions in this emerging field.

Building natural language inference (NLI) benchmarks that are both challenging for modern techniques, and free from shortcut biases is difficult.Chief among these biases is single sentence label leakage, where annotatorintroduced spurious correlations yield datasets where the logical relation between (premise, hypothesis) pairs can be accurately predicted from only a single sentence, something that should in principle be impossible.We demonstrate that despite efforts to reduce this leakage, it persists in modern datasets that have been introduced since its 2018 discovery.To enable future amelioration efforts, introduce a novel model-driven technique, the progressive evaluation of cluster outliers (PECO) which enables both the objective measurement of leakage, and the automated detection of subpopulations in the data which maximally exhibit it.

The recent LLMs like GPT-4 and PaLM-2 have made tremendous progress in solving fundamental math problems like GSM8K by achieving over 90% accuracy.However, their capabilities to solve more challenging math problems which require domain-specific knowledge (i.e.theorem) have yet to be investigated.In this paper, we introduce TheoremQA, the first theorem-driven question-answering dataset designed to evaluate AI models' capabilities to apply theorems to solve challenging science problems.TheoremQA is curated by domain experts containing 800 high-quality questions covering 350 theorems 1 from Math, Physics, EE&CS, and Finance.We evaluate a wide spectrum of 16 large language and code models with different prompting strategies like Chain-of-Thoughts and Program-of-Thoughts.We found that GPT-4's capabilities to solve these problems are unparalleled, achieving an accuracy of 51% with Program-of-Thoughts Prompting.All the existing open-sourced models are below 15%, barely surpassing the random-guess baseline.Given the diversity and broad coverage of TheoremQA, we believe it can be used as a better benchmark to evaluate LLMs' capabilities to solve challenging science problems.

The field of text-to-image (T2I) generation has garnered significant attention both within the research community and among everyday users.Despite the advancements of T2I models, a common issue encountered by users is the need for repetitive editing of input prompts in order to receive a satisfactory image, which is time-consuming and labor-intensive.Given the demonstrated text generation power of largescale language models, such as GPT-k, we investigate the potential of utilizing such models to improve the prompt editing process for T2I generation.We conduct a series of experiments to compare the common edits made by humans and GPT-k, evaluate the performance of GPT-k in prompting T2I, and examine factors that may influence this process.We found that GPT-k models focus more on inserting modifiers while humans tend to replace words and phrases, which includes changes to the subject matter.Experimental results show that GPT-k are more effective in adjusting modifiers rather than predicting spontaneous changes in the primary subject matters.Adopting the edit suggested by GPT-k models may reduce the percentage of remaining edits by 20-30%. 1 Our experiments are conducted upon StableDiffusion since it is a wide-adopted open-source large text-to-image generative model with SoTA performance.

In recent years, pre-trained large language models (LLMs) have demonstrated remarkable efficiency in achieving an inference-time few-shot learning capability known as in-context learning. However, existing literature has highlighted the sensitivity of this capability to the selection of few-shot demonstrations. Current understandings of the underlying mechanisms by which this capability arises from regular language model pretraining objectives remain disconnected from the real-world LLMs. This study aims to examine the in-context learning phenomenon through a Bayesian lens, viewing real-world LLMs as latent variable models. On this premise, we propose an algorithm to select optimal demonstrations from a set of annotated data with a small LM, and then directly generalize the selected demonstrations to larger LMs. We demonstrate significant improvement over baselines, averaged over eight GPT models on eight real-world text classification datasets. We also demonstrate the real-world usefulness of our algorithm on GSM8K, a math word problem dataset. Our empirical findings support our hypothesis that LLMs implicitly infer a latent variable containing task information.

Broader disclosive transparency-truth and clarity in communication regarding the function of AI systems-is widely considered desirable.Unfortunately, it is a nebulous concept, difficult to both define and quantify.This is problematic, as previous work has demonstrated possible trade-offs and negative consequences to disclosive transparency, such as a confusion effect, where "too much information" clouds a reader's understanding of what a system description means.Disclosive transparency's subjective nature has rendered deep study into these problems and their remedies difficult.To improve this state of affairs, We introduce neural language model-based probabilistic metrics to directly model disclosive transparency, and demonstrate that they correlate with user and expert opinions of system transparency, making them a valid objective proxy.Finally, we demonstrate the use of these metrics in a pilot study quantifying the relationships between transparency, confusion, and user perceptions in a corpus of real NLP system descriptions.

Recently developed large pre-trained language models, e.g., BERT, have achieved remarkable performance in many downstream natural language processing applications. These pre-trained language models often contain hundreds of millions of parameters and suffer from high computation and latency in real-world applications. It is desirable to reduce the computation overhead of the models for fast training and inference while keeping the model performance in downstream applications. Several lines of work utilize knowledge distillation to compress the teacher model to a smaller student model. However, they usually discard the teacher's knowledge when in inference. Differently, in this paper, we propose RemERT to leverage the knowledge learned from the teacher, i.e., facilitating the pre-computed BERT representation on the reference sample and compressing BERT into a smaller student model. To guarantee our proposal, we provide theoretical justification on the loss function and the usage of reference samples. Significantly, the theoretical result shows that including the pre-computed teacher's representations on the reference samples indeed increases the mutual information in learning the student model. Finally, we conduct the empirical evaluation and show that our RemERT can beat the vanilla TinyBERT over 8.1 % and achieves more than 94% of the performance of$\mathbf{BERT}_{\mathbf{BASE}}$on the GLUE benchmark. Meanwhile, RemERT is$\mathbf{7.4x}$smaller and$\mathbf{9.5x}$faster on inference than$\mathbf{BERT}_{\mathbf{BASE}}$.

Although deep learning models have driven state-of-the-art performance on a wide array of tasks, they are prone to spurious correlations that should not be learned as predictive clues. To mitigate this problem, we propose a causality-based training framework to reduce the spurious correlations caused by observed confounders. We give theoretical analysis on the underlying general Structural Causal Model (SCM) and propose to perform Maximum Likelihood Estimation (MLE) on the interventional distribution instead of the observational distribution, namely Counterfactual Maximum Likelihood Estimation (CMLE). As the interventional distribution, in general, is hidden from the observational data, we then derive two different upper bounds of the expected negative log-likelihood and propose two general algorithms, Implicit CMLE and Explicit CMLE, for causal predictions of deep learning models using observational data. We conduct experiments on both simulated data and two real-world tasks: Natural Language Inference (NLI) and Image Captioning. The results show that CMLE methods outperform the regular MLE method in terms of out-of-domain generalization performance and reducing spurious correlations, while maintaining comparable performance on the regular evaluations.

Topic modeling utilizing neural variational inference has shown promising results recently. Unlike traditional Bayesian topic models, neural topic models use deep neural network to approximate the intractable marginal distribution and thus gain strong generalisation ability. However, neural topic models are unsupervised model. Directly using the document-specific topic proportions in downstream prediction tasks could lead to sub-optimal performance. This paper presents Topic Attention Model (TAM), a supervised neural topic model that integrates an attention recurrent neural network (RNN) model. We design a novel way to utilize document-specific topic proportions and global topic vectors learned from neural topic model in the attention mechanism. We also develop backpropagation inference method that allows for joint model optimisation. Experimental results on three public datasets show that TAM not only significantly improves supervised learning tasks, including classification and regression, but also achieves lower perplexity for the document modeling.