Diversity

On-policy reinforcement learning (RL) for language model post-training suffers from a fundamental tension: as training progresses, policy entropy collapses and sampling diversity diminishes, causing the model to ``forget'' its own earlier exploratory capacity. While off-policy data can restore diversity, existing methods mix entire trajectories at the sequence level, introducing severe policy mismatch and training instability. We argue that the core question is not \emph{whether} to use off-policy data, but \emph{where} in the sequence it should appear. Based on this insight, we propose \textbf{Thinking Seeds}, a token-level mix-policy framework that uses the model's own historical checkpoints as off-policy prefixes, providing diverse starting points for reasoning, while the critical continuation is generated on-policy to preserve gradient quality. Through token-level importance ratios, Thinking Seeds effectively leverages historical diversity without compromising training stability. Extensive experiments across models and mathematical reasoning benchmarks demonstrate that Thinking Seeds consistently outperforms standard on-policy training and existing off-policy extensions. Our analysis reveals that the method maintains higher effective entropy, reduces gradient loss from clipping, and expands the explorable solution space, clarifying how position-aware mix-policy modeling improves both exploration and final performance in LLM RL.

RL-trained Lean theorem provers mode-collapse at inference time: on miniF2F-test with DeepSeek-Prover-V1.5-RL, doubling the i.i.d.\ sampling budget from $k{=}32$ to $k{=}64$ produces zero additional solved theorems (42/244 in both cases). A fixed schedule of 15 tactic skeletons breaks this plateau and recovers a $+45$ relative improvement at $k{=}16$ (mean $Δ= +12.3 \pm 4.2$ theorems across $n{=}3$ seeds, sign preserved in every seed). A controlled diversity ablation rules out the prompt-diversity confound: tactic skeletons help, paraphrases match the baseline, and irrelevant Lean comments actively degrade. A leave-one-out formalization-difficulty stratification reveals a structural-content gradient across the three perturbations. The phenomenon is RL-specific: V1.5-Base proves zero theorems regardless of intervention, identifying RL as the stage that creates the proof capability which subsequently collapses; extending to two additional 7B Lean provers, RL-trained DeepSeek-Prover-V2-7B contributes $+3$ frontier solves no i.i.d. baseline can reach despite a flat aggregate, while SFT-trained Goedel-Prover does not ( $-10.0$ $\pm 4.4$ theorems, $n{=}3$, sign preserved every seed). Inference-time structural diversity is a cheap, complementary axis for RL-trained provers, orthogonal to scaling model size or training compute.

Federal research funding shapes the direction, diversity, and impact of the US scientific enterprise. Large language models (LLMs) are rapidly diffusing into scientific practice, holding substantial promise while raising widespread concerns. Despite growing attention to AI use in scientific writing and evaluation, little is known about how the rise of LLMs is reshaping the public funding landscape. Here, we examine LLM involvement at key stages of the federal funding pipeline by combining two complementary data sources: confidential National Science Foundation (NSF) and National Institutes of Health (NIH) proposal submissions from two large US R1 universities, including funded, unfunded, and pending proposals, and the full population of publicly released NSF and NIH awards. We find that LLM use rises sharply beginning in 2023 and exhibits a bimodal distribution, indicating a clear split between minimal and substantive use. Across both private submissions and public awards, higher LLM involvement is consistently associated with lower semantic distinctiveness, positioning projects closer to recently funded work within the same agency. The consequences of this shift are agency-dependent. LLM use is positively associated with proposal success and higher early-stage publication output at NIH, whereas no comparable associations are observed at NSF. Notably, the productivity gains at NIH are concentrated in non-hit papers rather than the most highly cited work. Together, these findings provide large-scale evidence that the rise of LLMs is reshaping how scientific ideas are positioned, selected, and translated into publicly funded research, with implications for portfolio governance, research diversity, and the long-run impact of science.

With the rising popularity of interdisciplinary work and increasing institutional incentives in this direction, there is a growing need to understand how resulting publications incorporate ideas from multiple disciplines. Existing computational approaches, such as affiliation diversity, keywords, and citation patterns, do not account for how individual citations are used to advance the citing work. Although, in line with addressing this gap, prior studies have proposed taxonomies to classify citation purpose, these frameworks are not well-suited to interdisciplinary research and do not provide quantitative measures of citation engagement quality. To address these limitations, we propose a framework for the evaluation of citation engagement in interdisciplinary Natural Language Processing (NLP) publications. Our approach introduces a citation purpose taxonomy tailored to interdisciplinary work, supported by an annotation study. We demonstrate the utility of this framework through a thorough analysis of publications at the intersection of NLP and Computational Social Science.

Image Quality Assessment (IQA) is a long-standing problem in computer vision. Previous methods typically focus on predicting numerical scores without explanation or providing low-level descriptions lacking precise scores. Recent reasoning-based vision language models (VLMs) have shown strong potential for IQA by jointly generating quality descriptions and scores. However, existing VLM-based IQA methods often suffer from unreliable reasoning due to their limited capability of integrating visual and textual cues. In this work, we introduce Zoom-IQA, a VLM-based IQA model to explicitly emulate key cognitive behaviors: uncertainty awareness, region reasoning, and iterative refinement. Specifically, we present a two-stage training pipeline: 1) supervised fine-tuning (SFT) on our Grounded-Rationale-IQA (GR-IQA) dataset to teach the model to ground its assessments in key regions, and 2) reinforcement learning (RL) for dynamic policy exploration, stabilized by our KL-Coverage regularizer to prevent reasoning and scoring diversity collapse, with a Progressive Re-sampling Strategy for mitigating annotation bias. Extensive experiments show that Zoom-IQA achieves improved robustness, explainability, and generalization. The application to downstream tasks, such as image restoration, further demonstrates the effectiveness of Zoom-IQA.

Generative AI models reproduce the human biases in their training data and further amplify them through mechanisms such as mode collapse. The loss of diversity produces homogenization, which not only harms the minoritized but impoverishes everyone. We argue homogenization should be a central concern in AI safety. To meaningfully characterize homogenization in Large Language Models (LLMs), we introduce a framework that allows stakeholders to encode their context and value system. We illustrate our approach with an experiment that surfaces gender bias in an LLM (Claude 3.5 Haiku) on an open-ended story prompt. Building from queer theory, we formalize homogenization in terms of normativity. Borrowing language from feminist theory, we introduce the concept of xeno-reproduction as a class of tasks for mitigating homogenization by promoting diversity. Our work opens a collaborative line of research that seeks to understand and advance diversity in AI.

Existing stylized motion generation models have shown their remarkable ability to understand specific style information from the style motion, and insert it into the content motion. However, capturing intra-style diversity, where a single style should correspond to diverse motion variations, remains a significant challenge. In this paper, we propose a clustering-based framework, ClusterStyle, to address this limitation. Instead of learning an unstructured embedding from each style motion, we leverage a set of prototypes to effectively model diverse style patterns across motions belonging to the same style category. We consider two types of style diversity: global-level diversity among style motions of the same category, and local-level diversity within the temporal dynamics of motion sequences. These components jointly shape two structured style embedding spaces, i.e., global and local, optimized via alignment with non-learnable prototype anchors. Furthermore, we augment the pretrained text-to-motion generation model with the Stylistic Modulation Adapter (SMA) to integrate the style features. Extensive experiments demonstrate that our approach outperforms existing state-of-the-art models in stylized motion generation and motion style transfer.

We introduce DiverseVAR, a framework that enhances the diversity of text-conditioned visual autoregressive models (VAR) at test time without requiring retraining, fine-tuning, or substantial computational overhead. While VAR models have recently emerged as strong competitors to diffusion and flow models for image generation, they suffer from a critical limitation in diversity, often producing nearly identical images even for simple prompts. This issue has largely gone unnoticed amid the predominant focus on image quality. We address this limitation at test time in two stages. First, inspired by diversity enhancement techniques in diffusion models, we propose injecting noise into the text embedding. This introduces a trade-off between diversity and image quality: as diversity increases, the image quality sharply declines. To preserve quality, we propose scale-travel: a novel latent refinement technique inspired by time-travel strategies in diffusion models. Specifically, we use a multi-scale autoencoder to extract coarse-scale tokens that enable us to resume generation at intermediate stages. Extensive experiments show that combining text-embedding noise injection with our scale-travel refinement significantly enhances diversity while minimizing image-quality degradation, achieving a new Pareto frontier in the diversity-quality trade-off. Project page: https://diverse-var.github.io

Image diversity remains a fundamental challenge for text-to-image diffusion models. Low-diversity generation often leads to repetitive outputs, increasing sampling redundancy and hindering both creative exploration and downstream applications. A key factor is the tendency of diffusion models to collapse toward strong modes in the learned distribution. Existing attempts to improve diversity, such as steering-based guidance, often introduce distortions that degrade image quality. To address this issue, we propose Token-Prompt Embedding Space Optimization (TPSO), a training-free and model-agnostic module. TPSO introduces learnable parameters to explore underrepresented regions of the token embedding space, reducing the tendency to repeatedly sample from strong modes of the distribution. Meanwhile, a prompt-level semantic constraint regulates distribution shifts, preventing quality degradation while preserving semantic fidelity. Extensive experiments on MS-COCO across three representative diffusion backbones demonstrate that TPSO substantially improves diversity, boosting performance from 1.10 to 4.18, while maintaining image quality with only a modest inference-time overhead of 3.6% to 8.9%. Code is available at: https://github.com/Open-Debin/TPSO.

Flow matching models effectively represent complex distributions, yet estimating expectations of functions of their outputs remains challenging under limited sampling budgets. Independent sampling often yields high-variance estimates, especially when rare but high-impact outcomes dominate the expectation. We propose a non-IID sampling framework that jointly draws multiple samples to cover diverse, salient regions of a flow matching model's generative distribution. To balance diversity and quality, we introduce a score-based regularization for the diversity mechanism (SR), which uses the score function, i.e., the gradient of the log probability, to ensure samples are pushed apart within high-density regions of the data manifold, mitigating off-manifold drift. To enable unbiased estimation when desired, we further develop an approach for importance weighting of non-IID flow samples by learning a residual velocity field that reproduces the marginal distribution of the non-IID samples and by evolving importance weights along trajectories. Empirically, our method produces diverse, high-quality samples and accurate importance-weight estimates and debiased expectation estimates, advancing the reliable characterization of flow matching model outputs.

Integration of Large Language Models with search/retrieval engines has become ubiquitous, yet these systems harbor a critical vulnerability that undermines their reliability. We present the first systematic investigation of "chameleon behavior" in LLMs: their alarming tendency to shift stances when presented with contradictory questions in multi-turn conversations (especially in search-enabled LLMs). Through our novel Chameleon Benchmark Dataset, comprising 17,770 carefully crafted question-answer pairs across 1,180 multi-turn conversations spanning 12 controversial domains, we expose fundamental flaws in state-of-the-art systems. We introduce two theoretically grounded metrics: the Chameleon Score (0-1) that quantifies stance instability, and Source Re-use Rate (0-1) that measures knowledge diversity. Our rigorous evaluation of Llama-4-Maverick, GPT-4o-mini, and Gemini-2.5-Flash reveals consistent failures: all models exhibit severe chameleon behavior (scores 0.391-0.511), with GPT-4o-mini showing the worst performance. Crucially, small across-temperature variance (less than 0.004) suggests the effect is not a sampling artifact. Our analysis uncovers the mechanism: strong correlations between source re-use rate and confidence (r=0.627) and stance changes (r=0.429) are statistically significant (p less than 0.05), indicating that limited knowledge diversity makes models pathologically deferential to query framing. These findings highlight the need for comprehensive consistency evaluation before deploying LLMs in healthcare, legal, and financial systems where maintaining coherent positions across interactions is critical for reliable decision support.

Post-training alignment often reduces LLM diversity, leading to a phenomenon known as mode collapse. Unlike prior work that attributes this effect to algorithmic limitations, we identify a fundamental, pervasive data-level driver: typicality bias in preference data, whereby annotators systematically favor familiar text as a result of well-established findings in cognitive psychology. We formalize this bias theoretically, verify it on preference datasets empirically, and show that it plays a central role in mode collapse. Motivated by this analysis, we introduce Verbalized Sampling, a simple, training-free prompting strategy to circumvent mode collapse. VS prompts the model to verbalize a probability distribution over a set of responses (e.g., "Generate 5 jokes about coffee and their corresponding probabilities"). Comprehensive experiments show that VS significantly improves performance across creative writing (poems, stories, jokes), dialogue simulation, open-ended QA, and synthetic data generation, without sacrificing factual accuracy and safety. For instance, in creative writing, VS increases diversity by 1.6-2.1x over direct prompting. We further observe an emergent trend that more capable models benefit more from VS. In sum, our work provides a new data-centric perspective on mode collapse and a practical inference-time remedy that helps unlock pre-trained generative diversity.

As LLMs rapidly saturate existing benchmarks, automated benchmark creation using LLMs (LLM as a benchmark) where a model generates test inputs (LLM as a testset) and evaluates outputs (LLM as an evaluator) has gained traction as a cheap alternative to human curation. We show that this paradigm has a fundamental problem: LLM-generated benchmarks systematically favor the model that created them. Using machine translation as our primary testbed, we find that self bias arises from two compounding sources, LLM as a testset and LLM as an evaluator, and their combination amplifies the effect. Crucially, even when test data is generated with explicit diversity controls, each modelś implicit stylistic tendencies produce homogeneous, model-specific outputs that inflate its own scores. Increasing source text diversity, using our proposed diversity metric, partially mitigates this bias. Self bias is strong enough to cause each model to rank itself first, overriding the peer consensus ordering. We confirm that the phenomenon extends to open-ended generation on the Chatbot Arena task.

Magnitude, obtained as a special case of Euler characteristic of enriched category, represents a sense of the size of metric spaces and is related to classical notions such as cardinality, dimension, and volume. While the studies have explained the meaning of magnitude from various perspectives, continuity also gives a valuable view of magnitude. Based on established results about continuity of magnitude and maximum diversity, this article focuses on continuity of weighting, a distribution whose totality is magnitude, and its variation corresponding to maximum diversity. Meanwhile, recent studies also illuminated the connection between magnitude and data analysis by applying magnitude theory to point clouds representing the data or the set of model parameters. This article will also provide an application for time series analysis by introducing a new kind of invariants of periodic time series, where the invariance follows directly from the continuity results. As a use-case, a simple machine learning experiment is conducted with real-world data, in which the suggested invariants improved the performance.

Diffusion-based text-to-image models are increasingly used for urban analysis and scenario generation, but their geographic knowledge and representational biases remain poorly understood. We evaluate FLUX 1-schnell and Stable Diffusion 3.5-Large in the United States by generating 150 street-view images for each state, each state capital, and a generic "USA" prompt. Images are embedded with DINO-v2 ViT-S/14 and compared with Fréchet Inception Distance (FID). Pairwise FID clustering shows that geographically proximate states and capitals often group together, indicating implicit geographic structure. However, the generic ``USA'' prompt collapses this diversity into a metropolitan stereotype: frontier, desert, tropical, rural, and small-city environments are underrepresented or distant in FID space. These results show that diffusion models can encode fine-grained geography while still reproducing narrow national-scale visual stereotypes.

Vision-Language Models (VLMs) have demonstrated exceptional performance in various multi-modal tasks. Recently, there has been an increasing interest in improving the personalization capabilities of VLMs. To better integrate user-provided concepts into VLMs, many methods use positive and negative samples to fine-tune these models. However, the scarcity of user-provided positive samples and the low quality of retrieved negative samples pose challenges for existing techniques. To reveal the relationship between sample and model performance, we systematically investigate the amount and diversity impact of positive and negative samples (easy and hard) on VLM personalization tasks. Based on the detailed analysis, we introduce Concept-as-Tree (CaT), which represents a concept as a tree structure, thereby enabling the data generation of positive and negative samples with varying difficulty and diversity, and can be easily extended to multi-concept scenarios. With a well-designed data filtering strategy, our CaT framework can ensure the quality of generated data, constituting a powerful pipeline. We perform thorough experiments with various VLM personalization baselines to assess the effectiveness of the pipeline, alleviating the lack of positive samples and the low quality of negative samples. Our results demonstrate that CaT equipped with the proposed data filter significantly enhances the capabilities of VLMs across personalization benchmarks. To the best of our knowledge, this work is the first controllable synthetic data pipeline for VLM personalization.

Generative Pretrained Transformers (GPTs) are foundational Large Language Models (LLMs) for text generation. However, individual LLMs often produce inconsistent outputs and exhibit biases, limiting their representation of diverse language patterns. The closed-source nature of many powerful LLMs further restricts industry applications due to data privacy concerns. Inspired by successes in text generation, LLM ensemble techniques are now increasingly explored for code generation. This article reviews these emerging ensemble approaches to enhance understanding, encourage further research, and promote practical implementation in both text and code generation. We categorize LLM ensembles into seven main methods - weight merging, knowledge fusion, mixture-of-experts, reward ensemble, output ensemble, routing, and cascading - analyzing capabilities of those approaches. Our findings highlight key benefits such as improved diversity representation, enhanced output quality, and greater application flexibility. These insights aid model selection for real-world tasks and crucially, lay groundwork for extending ensemble strategies to multimodal LLMs.

How diverse are the outputs of large language models when diversity is desired? We examine the diversity of responses of several language models to questions with multiple possible answers, comparing them with human responses. Our findings suggest that models' responses are highly concentrated, reflecting narrow, mainstream outputs, in comparison to humans, whose responses exhibit a much longer-tail. We examine three simple and practical ways to increase output diversity: 1) increasing generation randomness via temperature sampling; 2) prompting models to answer from diverse perspectives using a single prompt; 3) aggregating outputs from several models. We find that these interventions, especially when combined, can substantially increase output diversity, although single-model outputs generally remain less diverse than the human baseline. We discuss potential implications of these findings for future work in AI policy and governance that wishes to preserve cultural diversity, an essential building block of a democratic social fabric.