We show that across architecture (Transformer vs. Mamba vs. RWKV), training dataset (OpenWebText vs. The Pile), and scale (14 million parameters to 12 billion parameters), autoregressive language models exhibit highly consistent patterns of change in their behavior over the course of pretraining. Based on our analysis of over 1,400 language model checkpoints on over 110,000 tokens of English, we find that up to 98% of the variance in language model behavior at the word level can be explained by three simple heuristics: the unigram probability (frequency) of a given word, the $n$-gram probability of the word, and the semantic similarity between the word and its context. Furthermore, we see consistent behavioral phases in all language models, with their predicted probabilities for words overfitting to those words' $n$-gram probabilities for increasing $n$ over the course of training. Taken together, these results suggest that learning in neural language models may follow a similar trajectory irrespective of model details.

Despite their impressive performance, vision-language models (VLMs) still struggle on culturally situated inputs. To understand how VLMs process culturally grounded information, we study the presence of culture-sensitive neurons, i.e. neurons whose activations show preferential sensitivity to inputs associated with particular cultural contexts. We examine whether such neurons are important for culturally diverse visual question answering and where they are located. Using the CVQA benchmark, we identify neurons of culture selectivity and perform causal tests by deactivating the neurons flagged by different identification methods. Experiments on three VLMs across 25 cultural groups demonstrate the existence of neurons whose ablation disproportionately harms performance on questions about the corresponding cultures, while having minimal effects on others. Moreover, we propose a new margin-based selector - Contrastive Activation Selection (CAS), and show that it outperforms existing probability- and entropy-based methods in identifying culture-sensitive neurons. Finally, our layer-wise analyses reveals that such neurons tend to cluster in certain decoder layers. Overall, our findings shed new light on the internal organization of multimodal representations.

Large Language Models have demonstrated strong performance on many established reasoning benchmarks. However, these benchmarks primarily evaluate structured skills like quantitative problem-solving, leaving a gap in assessing flexible, multifaceted reasoning abilities that are central to human intelligence. These abilities require integrating logical deduction with spatial awareness and constraint satisfaction, which current evaluations do not measure well. To address this, we introduce RiddleBench, a benchmark of 1,737 challenging puzzles in English designed to probe these core reasoning capabilities. Evaluation of state-of-the-art models on RiddleBench shows fundamental weaknesses. Even top proprietary models like Gemini 2.5 Pro, o3, and Claude 4 Sonnet achieve accuracy just above 60% (60.30%, 63.37%, and 63.16%). Analysis further reveals deep failures, including hallucination cascades (accepting flawed reasoning from other models) and poor self-correction due to a strong self-confirmation bias. Their reasoning is also fragile, with performance degrading significantly when constraints are reordered or irrelevant information is introduced. RiddleBench functions as a diagnostic tool for these issues and as a resource for guiding the development of more robust and reliable language models.

Large language models (LLMs) have demonstrated significant potential to accelerate scientific discovery as valuable tools for analyzing data, generating hypotheses, and supporting innovative approaches in various scientific fields. In this work, we investigate how LLMs can handle the transition from conceptual research ideas to well-structured research plans. Effective research planning not only supports scientists in advancing their research but also represents a crucial capability for the development of autonomous research agents. Despite its importance, the field lacks a systematic understanding of LLMs' research planning capability. To rigorously measure this capability, we introduce the Idea2Plan task and Idea2Plan Bench, a benchmark built from 200 ICML 2025 Spotlight and Oral papers released after major LLM training cutoffs. Each benchmark instance includes a research idea and a grading rubric capturing the key components of valid plans. We further propose Idea2Plan JudgeEval, a complementary benchmark to assess the reliability of LLM-based judges against expert annotations. Experimental results show that GPT-5 and GPT-5-mini achieve the strongest performance on the benchmark, though substantial headroom remains for future improvement. Our study provides new insights into LLMs' capability for research planning and lay the groundwork for future progress.

Benchmarks for out-of-distribution (OOD) generalization frequently show a strong positive correlation between in-distribution (ID) and OOD accuracy across models, termed "accuracy-on-the-line." This pattern is often taken to imply that spurious correlations - correlations that improve ID but reduce OOD performance - are rare in practice. We find that this positive correlation is often an artifact of aggregating heterogeneous OOD examples. Using a simple gradient-based method, OODSelect, we identify semantically coherent OOD subsets where accuracy on the line does not hold. Across widely used distribution shift benchmarks, the OODSelect uncovers subsets, sometimes over half of the standard OOD set, where higher ID accuracy predicts lower OOD accuracy. Our findings indicate that aggregate metrics can obscure important failure modes of OOD robustness. We release code and the identified subsets to facilitate further research.

We introduce MiRAGE, an evaluation framework for retrieval-augmented generation (RAG) from multimodal sources. As audiovisual media becomes a prevalent source of information online, it is essential for RAG systems to integrate information from these sources into generation. However, existing evaluations for RAG are text-centric, limiting their applicability to multimodal, reasoning intensive settings because they don't verify information against sources. MiRAGE is a claim-centric approach to multimodal RAG evaluation, consisting of InfoF1, evaluating factuality and information coverage, and CiteF1, measuring citation support and completeness. We show that MiRAGE, when applied by humans, strongly aligns with extrinsic quality judgments. We additionally introduce automatic variants of MiRAGE and three prominent TextRAG metrics -- ACLE, ARGUE, and RAGAS -- demonstrating the limitations of text-centric work and laying the groundwork for automatic evaluation. We release open-source implementations and outline how to assess multimodal RAG.

Grammar refers to the system of rules that governs the structural organization and the semantic relations among linguistic units such as sentences, phrases, and words within a given language. In natural language processing, there remains a notable scarcity of grammar focused evaluation protocols, a gap that is even more pronounced for low-resource languages. Moreover, the extent to which large language models genuinely comprehend grammatical structure, especially the mapping between syntactic structures and meanings, remains under debate. To investigate this issue, we propose a Grammar Book Guided evaluation pipeline intended to provide a systematic and generalizable framework for grammar evaluation consisting of four key stages, and in this work we take Luxembourgish as a case study. The results show a weak positive correlation between translation performance and grammatical understanding, indicating that strong translations do not necessarily imply deep grammatical competence. Larger models perform well overall due to their semantic strength but remain weak in morphology and syntax, struggling particularly with Minimal Pair tasks, while strong reasoning ability offers a promising way to enhance their grammatical understanding.

Using Reinforcement Learning with Verifiable Rewards (RLVR) to optimize Large Language Models (LLMs) can be conceptualized as progressively editing a query's `Reasoning Tree'. This process involves exploring nodes (tokens) and dynamically modifying the model's policy at each node. When combined with data scheduling, this process yields further gains in data efficiency and accuracy. However, existing RLVR data scheduling methods typically rely on path-based metrics to rank queries, overlooking the reasoning tree structures of these queries. In this paper, we introduce a novel metric, namely Reasoning Score (r-score), which measures the query's learning difficulty based on the structure of its reasoning tree. Based on the r-score, we propose the Reasoning Tree Schedule (Re-Schedule), a scheduling algorithm that constructs a curriculum progressing from structurally simple (high r-score) to complex (low r-score) queries. Experiments on six math-reasoning benchmarks show that Re-Schedule significantly improves average accuracy, achieving gains of up to 3.2%. These strong results validate our approach and demonstrate that a structural understanding of the reasoning tree provides a more powerful and principled foundation for RLVR data scheduling.

Large Language Models (LLMs) are powerful but often too slow and costly for real-world use during inference. Looped transformers save on parameters by reusing the same weights for multiple computational steps, or "loops." However, this approach has a major flaw: the loops run one after another, causing inference latency and memory requirements to increase with each added loop. This makes them impractical for fast applications. To solve this problem, we introduce the Parallel Loop Transformer (PLT). PLT is a new architecture that delivers the performance benefits of a deep, looped model but with the low latency of a standard, non-looped model. PLT works using two key techniques. First, Cross-Loop Parallelism (CLP) breaks the sequential dependency by computing different loops for different tokens at the same time, all within a single pass. Second, to prevent memory costs from growing, we use an Efficient Representation Enhancement strategy. This method shares the memory (KV cache) from the first loop with all other loops. It then uses a Gated Sliding-Window Attention (G-SWA) to combine this shared global information with local information, maintaining high accuracy. Our experiments show that PLT achieves the high accuracy of a traditional looped model but with almost no extra latency or memory cost compared to a standard transformer.

Introduction In April 2025, sixteen - year - old Adam Raine committed suicide . Over the course of several months, the teen confided his suicidal thoughts to Open AI's ChatGPT chatbot . ChatGPT is not designed or developed to provide therapy, but it did not respond to Adam's prompts with suggestions that he obtain professional help . Moreover, w hen Adam expressed concern that his parents would blame themselves if he died, ChatGPT reportedly responded, "That doesn't mean you owe them survival," and offered to help draft his suicide note. Adam's death was not the only example of chatbot misbehavior. OpenAI claims it doesn't permit ChatGPT "to generate hateful, harassing, violent, or adult content." In July 2025, a reporter documented ChatGPT providing users with detailed instructions for self - mutilation, murder, and satanic rituals. O penAI has also acknowledged that individuals can misuse its systems. But the company has taken some responsibility.