Understanding how opinions on different issues evolve together is essential for modeling collective intelligence, yet this remains underexplored due to the absence of standardized benchmarks. We introduce the concept of an opinion graph, where nodes represent social opinions on real-world events (e.g., presidential elections, stock predictions) and edges capture pairwise relationships between them. Building on this, we present OpinionBench, a continuously evolving benchmark derived from multiple live prediction markets, designed to evaluate whether large language models (LLMs) can uncover the latent structure within evolving social opinions. OpinionBench employs an scalable and automated graph construction and labeling pipeline, which assigns edge labels using time-series co-movement, semantic similarity, and metadata, followed by human validation. Experiments show that (1) LLMs consistently outperform baselines in identifying opinion correlations across domains, and (2) LLMs can recover the underlying graph structure through edge prediction. OpinionBench provides a challenging testbed for assessing LLMs’ ability to capture complex patterns of social opinion co-evolution.

Large language models (LLMs) are increasingly used to assist ideation in research, but evaluating the quality of LLM-generated research proposals remains difficult: novelty and soundness are hard to measure automatically, and large-scale human evaluation is costly. We propose a verifiable alternative by reframing proposal generation as a time-sliced scientific forecasting problem. Given a research question and inspiring papers available before a cutoff time, the model generates a structured proposal and is evaluated by whether it anticipates research directions that appear in papers published after the time. We operationalize this objective with the Future Alignment Score (FAS), computed via retrieval and LLM-based semantic scoring against a held-out future corpus. To train models, we build a time-consistent dataset of 21,835 paper occurrences across 3,642 instances from targets and their pre-cutoff citations, and synthesize reasoning traces that teach gap identification and inspiration borrowing. Across Llama-3.1 and Qwen2.5 models, future-aligned tuning improves future alignment over unaligned baselines (up to +10.6% overall FAS), and domain-expert human evaluation corroborates improved proposal quality. Finally, we demonstrate practical impact by implementing two model-generated proposals with a code agent, obtaining 4.17% accuracy gain on MATH from a new prompting strategy and consistent improvements for a novel model-merging method.

Large language models increasingly generate research ideas, yet judging their novelty or feasibility at generation time does not establish whether they anticipate subsequent work. We introduce IdeaForecastBench to evaluate research idea forecasting. Given a community’s literature up to a cutoff, a system produces up to five ranked ideas, which are evaluated against later papers. The benchmark comprises 624 rolling episodes across 52 topics, with a fixed retrieve-then-judge protocol and separately reported results from two judges. We compare five history-compression strategies across GPT-4.1, Qwen2.5-7B/14B, and Qwen3.5-9B, together with a learned Mode-Decomposition Forecaster (MDF). Under the primary GPT-4.1-mini judge, Summary improves on Direct in Hit@5 and Precision@5 across all four backbones. Qwen2.5 scores above GPT-4.1, whereas Qwen3.5 scores below it. An outcome-blind assessment finds that Qwen2.5 produces broader forecasts, but does not identify how much breadth contributes to its advantage. Threshold and judge diagnostics further clarify the limits of interpreting realization as precise anticipation. IdeaForecastBench provides a common task for studying which research ideas a community subsequently pursues and how reliably this outcome can be measured.

Reinforcement Learning (RL) has demonstrated significant potential in enhancing the capabilities of large language models (LLMs). However, the success of RL for LLMs heavily relies on human-curated datasets, which pose a burden on scalability and sustainability. Inspired by the success of the paradigm in games and Go, Self-Play methods are deployed to enhance LLM capabilities without human-annotated data. However, these methods primarily depend on a grounded environment for verifiable feedback (e.g., a Python interpreter or a game engine); extending them to general domains remains challenging. To address these challenges, we propose **Multi-Agent Evolve (MAE)**, a framework in which LLM can self-evolve across diverse tasks from mathematics to general knowledge. MAE implements asymmetric Self-Play by leveraging a triplet of interacting roles (*Proposer, Solver, Judge*) that are instantiated from a single LLM. *Proposer* generates challenging questions, *Solver* optimizes solutions, and *Judge* provides reward signals. These roles are jointly optimized via reinforcement learning, creating a closed-loop system where the model continuously challenges and improves itself. Experiments on Qwen2.5-3B-Instruct demonstrate that MAE achieves an average improvement of 4.86% across multiple benchmarks, significantly outperforming baselines. These results highlight MAE as a scalable, data-efficient method for enhancing LLMs’ general reasoning abilities with minimal reliance on human supervision.

Large language models’ reasoning abilities benefit from methods that organize their thought processes, such as chain-of-thought prompting, which employs a sequential structure to guide the reasoning process step-by-step. However, existing approaches focus primarily on organizing the sequence of thoughts, leaving structure in individual thought steps underexplored. To address this gap, we propose Table as Thought, a framework inspired by cognitive neuroscience theories on human thought. Table as Thought organizes reasoning within a tabular schema, where rows represent sequential thought steps and columns capture critical constraints and contextual information to enhance reasoning. The reasoning process iteratively populates the table until self-verification ensures completeness and correctness. Our experiments show that Table as Thought excels in planning tasks and demonstrates a strong potential for enhancing LLM performance in mathematical reasoning compared to unstructured thought baselines. This work provides a novel exploration of refining thought representation within LLMs, paving the way for advancements in reasoning and AI cognition.

Recent advancements in large language model (LLM) agents have significantly accelerated scientific discovery automation, yet concurrently raised critical ethical and safety concerns. To systematically address these challenges, we introduce SafeScientist, an innovative AI scientist framework explicitly designed to enhance safety and ethical responsibility in AI-driven scientific exploration. SafeScientist proactively refuses ethically inappropriate or high-risk tasks and rigorously emphasizes safety throughout the research process. To achieve comprehensive safety oversight, we integrate multiple defensive mechanisms, including prompt monitoring, agent-collaboration monitoring, tool-use monitoring, and an ethical reviewer component. Complementing SafeScientist, we propose SciSafetyBench , a novel benchmark specifically designed to evaluate AI safety in scientific contexts, comprising 240 high-risk scientific tasks across 6 domains, alongside 30 specially designed scientific tools and 120 tool-related risk tasks. Extensive experiments demonstrate that SafeScientist significantly improves safety performance by 35% compared to traditional AI scientist frameworks, without compromising scientific output quality. Additionally, we rigorously validate the robustness of our safety pipeline against diverse adversarial attack methods, further confirming the effectiveness of our integrated approach. The code and data will be available at https://github.com/ulab-uiuc/SafeScientist. Warning: this paper contains example data that may be offensive or harmful.

Automatic research with Large Language Models (LLMs) is rapidly gaining importance, driving the development of increasingly complex workflows involving multi-agent systems, planning, tool usage, code execution, and human-agent interaction to accelerate research processes. However, as more researchers and developers begin to use and build upon these tools and platforms, the complexity and difficulty of extending and maintaining such agentic workflows have become a significant challenge, particularly as algorithms and architectures continue to advance. To address this growing complexity, TinyScientist identifies the essential components of the automatic research workflow and proposes an interactive, extensible, and controllable framework that adapts easily to new tools and supports iterative growth. We provide an open-source codebase, an interactive web demonstration, and a PyPI Python package to make state-of-the-art auto-research pipelines broadly accessible to every researcher and developer.

When exposed to complex queries containing multiple conditions, today’s large language models (LLMs) tend to produce responses that only partially satisfy the query while neglecting certain conditions. We, therefore, introduce the concept of Intent Hallucination, a phenomenon where LLMs either omit (failing to address certain parts) or misinterpret (responding to invented query parts) elements of the given query, leading to responses misaligned with the original query. To systematically evaluate intent hallucination, we introduce FAITHQA, a novel benchmark for intent hallucination that contains 20,068 problems, covering both query-only and retrieval-augmented generation (RAG) setups with varying topics and difficulty. FAITHQA is the first hallucination benchmark that goes beyond factual verification, tailored to identify the fundamental cause of intent hallucination. By evaluating various LLMs on FAITHQA, we find that (1) intent hallucination is a common issue even for state-of-the-art models, and (2) such a phenomenon stems from omission or misinterpretation of LLMs. To facilitate future research, we introduce an automatic LLM generation evaluation metric, named INTENT CONSTRAINT, for detecting intent hallucination. Human evaluation results demonstrate that INTENT CONSTRAINT is closer to human performance for intent hallucination compared to baselines.

Evaluating Large Language Models (LLMs) requires effective methods to assess semantic consistency across multiple reversible transformations. Traditional self-consistency methods often fail to capture subtle semantic errors in multi-step tasks. We introduce ConsistencyChecker, a tree-based evaluation framework that measures LLMs’ ability to preserve semantic consistency during reversible transformation processes, sidestepping benchmark data contamination issues. Our approach constructs self-consistency trees where nodes represent text states after transformations (e.g., translation, code modification, paraphrasing) and edges represent pairs of opposite transformations. By analyzing semantic preservation between nodes at different tree depths, ConsistencyChecker quantifies model reliability without requiring manually annotated reference data. Experiments demonstrate that ConsistencyChecker reliably measures generalization abilities across models from 1.5B to 72B parameters. On translation tasks, GPT-4o Mini achieves the highest L3 consistency score of 98.0%. For code generation, Qwen 2.5 32B leads with 85.1% semantic consistency at L3. Results show Pearson correlation greater than 0.7 between our embedding-based scores and WMT 2024 rankings on 4 out of 5 shared language pairs, validating the method’s effectiveness for benchmarking LLM performance without constructing new datasets.

Recent advancements in artificial intelligence have led to the creation of highly capable large language models (LLMs) that can perform tasks in a human-like manner. However, LLMs exhibit only infant-level cognitive abilities in certain areas. One such area is the A-Not-B error, a phenomenon seen in infants where they repeat a previously rewarded behavior despite well-observed changed conditions. This highlights their lack of inhibitory control – the ability to stop a habitual or impulsive response. In our work, we design a text-based multi-choice QA scenario similar to the A-Not-B experimental settings to systematically test the inhibitory control abilities of LLMs. We found that state-of-the-art LLMs (like Llama3-8b) perform consistently well with in-context learning (ICL) but make errors and show a significant drop of as many as 83.3% in reasoning tasks when the context changes trivially. This suggests that LLMs only have inhibitory control abilities on par with human infants in this regard, often failing to suppress the previously established response pattern during ICL.

Advances in multimodal models have greatly improved how interactions relevant to various tasks are modeled. Today’s multimodal models mainly focus on the correspondence between images and text, using this for tasks like image-text matching. However, this covers only a subset of real-world interactions. Novel interactions, such as sarcasm expressed through opposing spoken words and gestures or humor expressed through utterances and tone of voice, remain challenging. In this paper, we introduce an approach to enhance multimodal models, which we call Multimodal Mixtures of Experts (MMoE). The key idea in MMoE is to train separate expert models for each type of multimodal interaction, such as redundancy present in both modalities, uniqueness in one modality, or synergy that emerges when both modalities are fused. On a sarcasm detection task (MUStARD) and a humor detection task (URFUNNY), we obtain new state-of-the-art results. MMoE is also able to be applied to various types of models to gain improvement.

Humans learn social skills through both imitation and social interaction. This social learning process is largely understudied by existing research on building language agents. Motivated by this gap, we propose an interactive learning method, SOTOPIA-π, that improves the social intelligence of language agents. This method leverages behavior cloning and self-reinforcement based training on filtered social interaction data according to large language model (LLM) rating. We show that our training method allows a 7B LLM to reach the social goal completion ability of an expert model (GPT-4-based agent) without the loss of more generic abilities, such as the ability to answer knowledge-based questions. We also demonstrate that this training paradigm uncovers some weaknesses in standard evaluation and safety training paradigms that (1) LLM-based evaluation of social intelligence overestimates the abilities of the language agents trained specifically for social interaction, and that (2) despite not training for better safety or question answering (QA) ability, our methods improve the safety of language agents and maintain general QA ability on the MMLU benchmark.

The Transformer architecture is crucial for numerous AI models, but it still faces challenges in long-range language modeling. Though several specific transformer architectures have been designed to tackle issues of long-range dependencies, existing methods like Transformer-XL are plagued by a high percentage of ineffective memories. In this study, we present a plug-and-play strategy, known as TRAining-free Memory Selection (TRAMS), that selects tokens participating in attention calculation based on one simple metric. This strategy allows us to keep tokens that are likely to have a high attention score with the current queries and ignore the other ones. We have tested our approach on the word-level benchmark (WikiText-103) and the character-level benchmark (enwik8), and the results indicate an improvement without having additional training or adding additional parameters.

Sample-and-rank is a key decoding strategy for modern generation-based dialogue systems. It helps achieve diverse and high-quality responses by selecting an answer from a small pool of generated candidates. The current state-of-the-art ranking methods mainly use an encoding paradigm called Cross-Encoder, which separately encodes each context-candidate pair and ranks the candidates according to their fitness scores. However, Cross-Encoder repeatedly encodes the same lengthy context for each candidate, resulting in high computational costs. Poly-Encoder addresses the above problems by reducing the interaction between context and candidates, but with a price of performance drop. In this work, we develop a new paradigm called Uni-Encoder, that keeps the full attention over each pair as in Cross-Encoder while only encoding the context once, as in Poly-Encoder. Uni-Encoder encodes all the candidates with the context in one forward pass. We use the same positional embedding for all candidates to ensure they are treated equally and design a new attention mechanism to avoid confusion. Our Uni-Encoder can simulate other ranking paradigms using different attention and response concatenation methods. Extensive experiments show that our proposed paradigm achieves new state-of-the-art results on four benchmark datasets with high computational efficiency. For instance, it improves R10@1 by 2.9% with an approximately 4X faster inference speed on the Ubuntu V2 dataset.

Open-Domain Question Answering (ODQA) systems necessitate a reader model capable of generating answers by simultaneously referring to multiple passages. Although representative models like Fusion-in-Decoder (FiD) have been proposed to address this challenge, these systems can inadvertently rely on spurious features instead of genuine causal relationships between the question and the passages to generate answers. To counter this problem, we introduce the Rational Fusion-in-Decoder (RFiD) model. Our model leverages the encoders of FiD to differentiate between causal relationships and spurious features, subsequently guiding the decoder to generate answers informed by this discernment. Experimental results on two ODQA datasets, Natural Questions (NQ) and TriviaQA (TQ), demonstrate that our model surpasses previous methods, achieving improvements of up to 1.5 and 0.7 in Exact Match scores on NQ, and exhibits an enhanced ability to identify causal relationships.

Large language models (LLMs) have recently reached an impressive level of linguistic capability, prompting comparisons with human language skills. However, there have been relatively few systematic inquiries into the linguistic capabilities of the latest generation of LLMs, and those studies that do exist (i) ignore the remarkable ability of humans to generalize, (ii) focus only on English, and (iii) investigate syntax or semantics and overlook other capabilities that lie at the heart of human language, like morphology. Here, we close these gaps by conducting the first rigorous analysis of the morphological capabilities of ChatGPT in four typologically varied languages (specifically, English, German, Tamil, and Turkish). We apply a version of Berko’s (1958) wug test to ChatGPT, using novel, uncontaminated datasets for the four examined languages. We find that ChatGPT massively underperforms purpose-built systems, particularly in English. Overall, our results—through the lens of morphology—cast a new light on the linguistic capabilities of ChatGPT, suggesting that claims of human-like language skills are premature and misleading.