Output Diversity in Language Models

Memory serves as the core component of LLM-based agents, enabling personalized responses via extracting and using user preferences from past interactions. However, its use may reduce response diversity, thereby narrowing the range of information available to users. Existing evaluations rely largely on synthetic interactions and cover few memory systems, leaving the effects of memory on response diversity insufficiently evaluated in real-world scenarios. We introduce MemBubble, a benchmark of 1000 instances built on preferences extracted from real user-assistant conversations. To measure the extent to which memory concentrates response distributions, we construct test queries in scenarios distinct from user preferences while preserving topical relevance through shared preference categories. Across multiple LLMs and memory mechanisms, entropy decreases by 34.7% on average compared with directly providing the full interaction history, highlighting the important role of memory processing in diversity loss. We also explore simple user-side interventions and changes in information availability but find that these do not effectively close the diversity gap. Our findings highlight the need to consider preserving diversity and user access to information alongside improvements in personalization when designing and using agent memory systems.

While aligned large language models (LLMs) can successfully complete many creative tasks, their outputs tend to be highly self-similar. Though solutions to this problem of "mode collapse" have been proposed, comprehensive large-scale surveys of the phenomenon in specific textual genres are still lacking. We address this gap through an extensive treatment of LLM mode collapse in poetry. Across 15 poetic forms, 248 subject matter tags, and 16 open-weight models, we generate over 1.1 million poems. To our knowledge, this constitutes the largest corpus of LLM-generated poetry to date. At million-poem scale, we find that LLM poetic vocabularies and rhyming abilities are substantial, but severely underutilized. In other words, LLMs are "lazy" poets: despite possessing large poetic inventories, LLMs over-rely on a small number of words, opening phrases, and rhymes. Further, we find that lexical homogenization differs across models and poetic forms, noting observed rhyming vocabulary is correlated with model size.

Post-training is widely assumed to reduce the social biases language models inherit from pretraining, yet direct evidence for this assumption remains mixed. In this paper, we examine how gender disparities evolve across the language modeling pipeline. In particular, we study occupational gender bias in OLMo 3, a fully open language model family, across its post-training stages by prompting each checkpoint to generate open-ended narratives about synthetic personas given their occupation and education. We consider two settings, a gender-given setting in which the profile states the persona's gender explicitly and a gender-assumed setting in which it must be inferred by the model. We find that when gender is stated, post-training increases semantic diversity for male-coded profiles, whereas for female-coded profiles the increase is smaller after SFT and not significant after DPO and RLVR. When gender is assumed by the model, post-training affects narratives about female and male personas similarly. The model's own gender assumptions, however, show a consistent female-preferential skew relative to United States labor-force demographics, except for heavily male-dominated occupations. Every successive post-training stage monotonically increases the odds of attributing a female label over the base model. A provenance analysis of the supervised fine-tuning corpus finds only a modest 7.7% female-token surplus, too small to explain the size of these downstream effects. The project code is available under: https://github.com/manthangadhia/llm-lifecycle-bias.

Text generated by Large Language Models (LLMs) often remains limited in diversity compared to that produced by humans. Recent advances demonstrate the presence of inter- and intra-model collapse in open-ended generation of English text, raising concerns about the long-term homogenization of LLM, and more importantly, human thought. However, our understanding of whether this homogenization extends to languages beyond English is limited. In this work, we investigate the role of language on output diversity. Specifically, we aim to understand whether multilingual LLM-generated text is also homogeneous, and whether this homogeneity matches what we observe in humans. Across 23 open-weight models and 13 languages, we find that the artificial hivemind effect is pervasive but uneven in its extent. The intra-model effect holds in three different embedding spaces and under a lexical metric that uses no embedding model, so it is not an artifact of how diversity is measured. The size of the homogenization gap is associated with a model's competence in a language, measured on a held-out translation benchmark, robustly across embedding spaces. Thus, we demonstrate a performance-diversity tradeoff in multilingual LLMs.

We introduce Gacha Decoding, an inference-time method for eliciting diverse language model generations that scales with model capability. Across open-ended domains (in-the-wild chat, creative writing, planning for image generation, and protein design), Gacha Decoding significantly outperforms existing generation diversity approaches at equal quality (up to 2.4x Vendi over the next-best prior approach), reaching the same number of high-quality modes with over an order of magnitude fewer samples (11.0x) and discovering novel modes that no other approach surfaces. Our key insight is to treat diversity as an instruction-following problem: rather than relying on the LM's token entropy, we combine its instruction-following capability with randomness from an external RNG tool to scalably identify and realize distinct modes of the response space. This approach of "planning with dice" enables Gacha to invert the long-observed tension between diversity and model capability. As the underlying LM becomes a better instruction follower, diversity under Gacha Decoding consistently improves--even as its token entropy and diversity under prior approaches decline. Together, our results highlight that instruction following, rather than token entropy alone, can drive generation diversity.

Recent research argues that language models exhibit pronounced homogeneity in open-ended generation, framing such behavior as an Artificial Hivemind that poses a long-term threat to human creativity. We examine three of its central results. First, the flagship example is that model responses to "Write a metaphor involving time" collapse into two clusters. Visualization, spectral analysis, clustering, and language model labels all contradict this description. The labels record each response's vehicle, what it compares time to. Our responses and the original authors' own show one dominant vehicle plus a heavy tail of distinct minority vehicles. "Time" is one of our least diverse topics, so the example is a favorable case, not a representative one. Second, the paper measures homogeneity against an undemanding null: responses to unrelated prompts. Under a more demanding null (same-prompt responses expressing genuinely different ideas), 20%-32% of such pairs already exceed the paper's 0.8 convergence threshold. A residual effect survives this null. The paper's same-prompt pairs exceed 0.8 roughly two to three times as often as our different-idea pairs. Much of what the paper calls homogeneity is the shared geometry of answering the same prompt. The remaining measurements lack any null: no human baseline is collected, and the model-indistinguishability statistic has no null. Third, the paper concludes that inference-time interventions are inadequate for combating the Artificial Hivemind, writing that "more generalizable solutions are needed at the model training level." We show that this conclusion is unsupported in three ways, and that an inference-time intervention (prompting) reliably raises measured response diversity. We do not resolve whether the Artificial Hivemind is real. We show that the published evidence does not establish it.

We propose DAST (Diversifying Arithmetic Sampling with TokenTour), a method that increases the diversity of LLM outputs without any change to the marginal distribution and with negligible generation-time overhead (a few microseconds). We observe that token IDs are often arranged in a meaningless order and reassign them so that tokens with similar meanings appear consecutively. This can be done in advance in a few hundred seconds per model, and the resulting order can be reused for all subsequent generations. By combining this order with arithmetic sampling (or quasi-Monte Carlo methods), we make similar tokens less likely to be generated across runs while preserving the distribution. Our method not only produces qualitatively good ideas but also significantly improves performance on the downstream task of ProtoQA.

Language models often produce homogeneous responses to open-ended tasks; such homogeneity can spawn groupthink-the convergence of ideas toward a singular and potentially suboptimal decision. We formulate persona diversification as a set-level conditioning problem and study two orthogonal design choices: selecting versus generating personas, and space-filling versus frontier-seeking diversity. We instantiate this design space with four methods spanning coverage and dispersion subset selections, uniform-coverage sampling, and evolutionary persona generation. Evaluations on the Alternative Uses Task (AUT), Infinity-Chat, and Divergent Association Task (DAT) show the benefits of the proposed methods across tasks and creativity objectives. On AUT, evolutionary persona generation increases response diversity by 78.8%, originality by 26.1%, flexibility by 49.5%, and holistic creativity by 13.9% over task-only prompting, while maintaining 98.5% validity; on Infinity-Chat, it nearly doubles persona-induced response separation relative to random personas. Moreover, evolutionary personas compose with creativity-optimized prompting, further increasing its response diversity by 18.6% and creativity by 6.3%. These results establish persona-set geometry as a task-agnostic mechanism for eliciting divergent LLM outputs, and support persona diversification as a reusable complement to prompt optimization.

Recent work by Doshi and Hauser (2024), Bisbee et al. (2024), and Xie et al. (2026) raises concerns that outputs from large language models (LLMs) tend to be under-diverse: they repeat or resemble one another more often than responses from the population they are meant to represent, a phenomenon known as mode collapse. In this work, we show that whether mode-collapse, or its opposite, occurs depends on the specific model and dataset used. Further, with sufficient supervised fine-tuning (SFT) data, LLM output diversity converges toward that of the target distribution from which fine-tuning data are sampled. To quantify this comparison, we measure the probability that two responses sampled independently from the same fixed prompt coincide (collide), or their expected similarity under a kernel. We derive a bias-variance decomposition of the expected gap between the model's and target's collision probabilities, showing that SFT is not inherently biased toward mode collapse or its opposite: finite-sample SFT can leave a model either under- or over-dispersed, depending on the model and dataset. Finally, we show that the absolute gap is bounded by the square root of the Kullback-Leibler (KL) divergence from the target distribution to the model. Consequently, a model sufficiently close to optimal under population cross-entropy cannot exhibit arbitrarily miscalibrated diversity. We test the decomposition and the bound in three experiments: small transformers on synthetic languages, four LLMs fine-tuned on human surveys, and these LLMs fine-tuned on CodeNet, a dataset of human code solutions. More target data moves model diversity toward the human (or synthetic target) level in all experiments, consistent with our theoretical predictions. These results show that diversity miscalibration can arise from finite-sample error and shrink as SFT better approximates the target distribution.

Many LLM applications are most useful when they provide several candidate outputs for comparison, validation, or combination. Predominant evaluation settings, however, still focus on individual outputs or reduce multiple samples to a single success or selected answer. This can miss whether the outputs include several genuinely different useful results. We introduce VTC-Bench, a five-domain benchmark for this setting, together with Validated Task Coverage (VTC) as its core evaluation quantity. The benchmark is built from carefully selected real-data tasks where both output quality and task-relevant distinctness can be checked automatically and reproducibly, without model-based judges. VTC measures how many distinct useful results are obtained within $k$ attempts. Across multiple models and inference settings, the benchmark leads to different conclusions from conventional evaluation: configurations that look strongest from single-draw quality are not necessarily those with the best coverage, and simple measures of output variation do not reliably recover task-relevant coverage. These results show that finite candidate sets can be evaluated directly as objects of interest, revealing differences in model behavior that are not apparent from conventional per-output evaluation.

While post-training improves the capabilities of large language models (LLMs), it generally lowers their output diversity and creativity, negatively impacting tasks that explicitly require creativity (e.g., story generation) as well as those that require it implicitly, e.g., reinforcement learning (RL). We instead propose CreativeInstruct, a scalable instruction-tuning method that teaches LLMs to balance creative, base-model-like generations with the quality of post-trained models, by learning to inject special [StartCreativity] spans that bias generation toward creativity. Furthermore, we introduce a structural diversity metric based on graph edit distance, which captures narrative level variation missed by purely lexical and semantic metrics. On narrative generation, CreativeInstruct matches or exceeds the diversity of both multi-model baselines and distilled variants of their outputs, without sacrificing quality or requiring multiple models at inference time. These results are mirrored in our human evaluation, where we find that annotators rate CreativeInstruct generations as more creative than the post-trained LLMs' generations in 70.3% of cases. We also show the benefits of creative models as a substrate for RL: GRPO applied to a CreativeInstruct checkpoint improves by ~4% on AMC and ~5% points on MATH over the same training applied to the post-trained checkpoint.

While large language model outputs are frequently analysed as a collective super variety termed "AI language," this chapter argues that this perspective coexists with distinct, model-specific linguistic signatures akin to human idiolects. We analyse two datasets of LLM-generated texts on societal topics: a 2024 corpus of six models (Improta et al. 2024) and a newly generated 2026 corpus using the same prompts featuring six contemporary models. Our findings, utilising computational descriptors and stylometric principal component analysis reveal a generational shift between the style of the 2024 and 2026 cohorts, while demonstrating that each individual model maintains a unique linguistic profile. This multi-layered interplay is illustrated by contraction frequencies, which vary from over 1,200 to over 30,000 per million words within the same cohort of models (2026). Ultimately, we conclude that treating LLM output as idiolectal in nature provides a valuable framework with potential implications for research on variation and change, LLM-generated text detection, forensic linguistics and usage-based approaches to language.

Large Language Models (LLMs) have demonstrated remarkable potential for analogy making, a core cognitive capability that drives novelty and creativity. While prior research has extensively investigated the applications and underlying mechanisms of LLM-based analogy making, its output diversity remains largely unexplored, despite being essential for broadening cross-domain connections and fostering scientific innovation. In this work, we present a comprehensive evaluation of analogy diversity across ten state-of-the-art open- and closed-source LLMs. Our findings highlight a concerning issue of domain homogeneity, a prevalent tendency for LLMs to generate analogies from a narrow set of target domains, limiting both inter-query and intra-model diversity. Furthermore, our analysis reveals a fundamental trade-off in existing LLM diversity-enhancement methods: increasing output diversity often comes at the expense of output quality. Finally, our causal analysis of LLM information flow reveals substantial differences in the model-sensitive regions governing analogy diversity across LLMs, suggesting a potential mechanism for the observed diversity-quality trade-off. To our knowledge, this is among the first studies to systematically investigate output diversity in LLM-based analogy making.

When a language model must choose one answer from a large space of equally valid options, a format clause -- "Reply with JSON only" -- changes which answer it chooses. We re-run the One-Word Census (arXiv:2607.12796): 31 wide-answer-space category prompts asked of 44 models, now with the reply requested in JSON -- no schema enforcement, no constrained decoding, only the request. Convergence deepens sharply: on the unconstrained "Pick a word" prompt the modal answer rises from 41% to 64% of the pool and distinct answers fall from 52 to 36; mean answer-choice surprisal drops from 1.80 to 1.58 bits. The tax is progressive: six of 44 models move individually (BH-FDR q=.10), all toward the mode, led by the most distinctive models, while the conformist floor is immobile. It is a sharpener, not a re-indexer -- the plain-chat modal answer survives in 28 of 31 categories. Defaults are register-indexed: a within-run re-sample (n=20) finds JSON shifts 53% of a model's stable chat defaults, mostly back to the crowd, and installs defaults absent from chat (Claude Fable 5 answers "cerulean" for colour 0% of the time in chat, 100% in JSON). Full-battery controls reveal a register gradient: compression is significant and specific to the answer-delivery formats models are trained to speak (JSON -0.22 bits, p=.0002; XML -0.19, p=.002), absent for YAML and CSV, and reversed for an arbitrary bracket wrapper (+0.13, p=.009) -- weighing the mechanism toward tool-use post-training. Enforcing the schema at the decoder (response_format) compresses no further than the request (-0.03 bits): the collapse lives in the model's response to the register, not the decoder. Structured output is how software consumes language models, and that surface is served by a measurably more homogeneous model than the chat surface on which models are evaluated, compared, and chosen.

When a language model must choose one answer from a large space of equally valid options, which answer does it choose, and how often is it the answer every other model chooses? Asked to "pick a word," 105 language models from more than twenty labs chose serendipity 46% of the time. We measure this convergence, and each model's share in it, with 96 single-turn prompts that each name a category with many valid one-word answers ("Name a tree."), asked eight times per model and scored by exact match, with no embeddings and no judge. A model's answer-choice surprisal is the average -log2 probability of its answers under the pooled answers of all other models. In 28 of 96 categories a single answer takes at least 80% of all answers. The concentration does not depend on small or persona-tuned models: the 87 major-lab models are at least as concentrated as the full field. Lightly post-trained and persona-tuned models are the most divergent; heavily post-trained assistants from the major labs are the most conformist. Models that avoid the modal answer mostly land on the same runner-up. Within the major providers' lineages, release order shows no panel-wide trend once model tier is controlled; GPT, Gemini, Grok and Qwen become more conformist across releases, and Claude's generation-5 releases reverse. On open checkpoints of three post-training pipelines, supervised fine-tuning is the largest step toward the field's answers. Against human category-production norms, the field is more concentrated than people in 18 of 20 shared categories. All prompts, transcripts, and code are public.

Open-ended generation tasks often require a set of meaningfully different outputs, yet large language models often produce similar generations. Existing test-time diversity methods operate at different stages of generation with varying effectiveness, but it remains unclear what design choices lead to meaningful diversity in the output. We introduce a framework that characterizes test-time diverse generation methods by the diversity source introduced during generation and provide a transmission score for measuring how effectively variation in the source reaches the final output. Guided by this framework, we propose fully automated specification-level generation methods that first generate diverse intermediate specifications and then condition on them to produce final responses. Across five open-ended tasks and four backbone models, specification-level injection improves output diversity over test-time baselines while maintaining comparable quality. Our analysis shows that successful diversity injection depends on both the diversity of the sources and their transmission to the output, highlighting source design and source-to-output realization as two key levers for building more diverse generation systems.

Measuring the diversity of creative outputs is central to evaluating post-training mode collapse, comparing decoding strategies, and quantifying creative behavior in both AI and human writing. We propose a new approach to measuring diversity using in-context learning, of which the ``Decan'' metric, $D_{Ca_n} = C \times a_n$, is the working instance we evaluate: a per-byte score read off the per-token log-probabilities of a base model $θ$ in a single forward pass per permutation, with no embedding model, no reference corpus, and no human labels. This approach is grounded in information theory, makes use of language model in-context learning to detect a wide range of similarities between any number of inputs, and obviates the need to train a special-purpose model. The same pipeline scores AI samples and human-written response sets, with diversity treated as a property of (responses, prompt, scoring model). On Tevet and Berant's human-grounded McDiv benchmark, $D_{Ca_n}$ reaches OCA 0.846 on the McDiv prompt_gen set where it performs best, behind the strongest neural baseline reported in Tevet and Berant (SentBERT, 0.897). On the OLMo-2-7B post-training pipeline, $D_{Ca_n}$ drops monotonically across the base $\to$ SFT $\to$ DPO $\to$ RLVR stages, detecting the type of diversity loss that creative-writing applications care about.

Fine-tuning is often believed to reduce uncertainty and diversity in large language models, but existing analyses overlook output length, a key confounder, and therefore fail to capture how uncertainty is distributed across an entire generation rollout. To address this, we propose Canopy Entropy ($\mathrm{CE}^\star$), a measure that views language generation from a tree perspective, where ``canopy'' represents the space of all possible rollouts, making $\mathrm{CE}^\star$ naturally quantify the effective size of the generation space. $\mathrm{CE}^\star$ jointly captures uncertainty in both the output length $N$ and the generated sequence $Y_{1:N}$ -- indeed, we show that it equals to total Shannon entropy $H(N, Y_{1:N}\mid X)$, where $X$ denotes the prompt. This formulation yields interpretable metrics, including a length-entropy correlation term $ρ(N, r_N)$, where $r_N$ is the entropy rate, quantifying information conveyance efficiency by indicating whether longer outputs are more or less informative per token. Empirically, across tasks and model families, we find that fine-tuned models consistently exhibit stronger positive correlation $ρ(N, r_N)$, even when total entropy decreases. Furthermore, after controlling for model family, task, prompt, and output-length effects, we find that fine-tuning nearly triples the correlation strength between entropy rate and semantic diversity, suggesting that aligned models convert token uncertainty into semantic diversity more efficiently. Overall, these results demonstrate that fine-tuning does not simply reduce uncertainty, but fundamentally reorganizes it into more informative and semantically meaningful generations. Our code is available at https://github.com/WeiyiTian/canopy-entropy.

Many open-ended instructions have multiple valid answers that users can benefit from seeing, but post-training often narrows an LLM's output space toward a small set of canonical responses. We introduce REDIPO, an offline DPO data-construction pipeline for recovering distinct valid answer modes while preserving the alignment benefits of the instruct model. For each prompt, REDIPO samples responses from both base and instruct models, rewrites base-model responses with the instruct model, filters candidates for safety and instruction-following quality, and builds preference pairs that favor marginally diverse responses among candidates with similar instruction-following reward. Across Qwen3-4B, OLMo-3-7B, and LLaMA-3.1-8B, REDIPO improves NoveltyBench distinct_k by 134%, 33%, and 44% relative to the instruct checkpoints, while DivPO changes diversity by 0%, -6%, and -4% on the same models. These gains largely maintain MTBench, IFEval, and Arena-Hard performance, and reduce direct-category HarmBench attack success rate. Ablations show that marginal-diversity pair selection and base-response rewriting drive the diversity gains, while filtering and quality-bounded pairing help maintain alignment. Overall, our results show that diverse valid answers from base-model generations can be reintroduced through carefully constructed preference data while retaining the alignment benefits of post-training. We release our code and data at https://github.com/vsamuel2003/ReDiPO.

When large language models are fine-tuned to generate persona- or tone-conditioned responses, their output diversity is severely limited--a failure we term Cross-Style Collapse. We trace this collapse to the cross-entropy objective, which under shared representations tends to suppress diverse continuations. We propose Semantic Flow Regularization (SFR), a lightweight auxiliary objective that supervises the backbone with continuous sentence-encoder embeddings of future segments via conditional flow matching. The stochastic flow source preserves multi-modality by construction; the flow-matching head is discarded at inference, adding zero deployment cost. On a large-scale industrial dialogue dataset (Qwen3-32B, 9 personas), SFR improves output diversity, style fidelity, and response quality over SFT. We further validate on the public LiveCodeBench-v5 (Qwen2.5-Coder-7B-Instruct), where SFR consistently improves pass@k, confirming generality beyond stylized dialogue. A controlled comparison on MBPP reveals Multi-Token Prediction to be a degenerate special case of SFR.

Diversity is essential for language-model applications ranging from creative generation to scientific discovery, yet modern LLMs often collapse into a narrow subset of plausible outputs. While prior work has developed benchmarks for measuring this lack of diversity, less is known about how the step-by-step probability distributions at inference time cause the problem. We introduce a validity--diversity framework that attributes diversity collapse to how an LLM allocates probability mass across valid and invalid continuations during decoding. This framework decomposes the bottleneck into two complementary forms of miscalibration. First, order calibration: valid tokens are not reliably ranked above invalid tokens, so rank-based cutoff rules must trade off between recovering valid continuations and admitting invalid ones. Second, shape calibration: probability mass is overly concentrated only on few valid continuations while having a heavy-tail of mixed valid and invalid tokens, so maintaining high validity limits diversity. We formalize both mechanisms and show that local failures compound across decoding steps, producing strong sequence-level losses in diversity. Empirically, we develop controlled diagnostics for probing these bottlenecks, including tasks with exactly known valid sets and oracle cutoff baselines. Across 14 language models spanning multiple families and scales, we find that diversity collapse is not merely a limitation of particular sampling heuristics, but a consequence of order and shape miscalibration in the LLM distribution.

Generating diverse responses is crucial for test-time scaling of large language models (LLMs), yet standard stochastic sampling mostly yields surface-level lexical variation, limiting semantic exploration. In this paper, we propose Exploratory Sampling (ESamp), a decoding approach that explicitly encourages semantic diversity during generation. ESamp is motivated by the well-known observation that neural networks tend to make lower-error predictions on inputs similar to those encountered before, and incur higher prediction error on novel ones. Building on this property, we train a lightweight Distiller at test time to predict deep-layer hidden representations of the LLM from its shallow-layer representations to model the LLM's depth-wise representation transitions. During decoding, the Distiller continuously adapts to the mappings induced by the current generation context. ESamp uses the prediction error as a novelty signal to reweight candidate token extensions conditioned on the current prefix, thereby biasing decoding toward less-explored semantic patterns. ESamp is implemented with an asynchronous training--inference pipeline, with less than 5% worst case overhead (1.2% in the optimized release). Empirical results show that ESamp significantly boosts the Pass@k efficiency of reasoning models, showing superior or comparable performance to strong stochastic and heuristic baselines. Notably, ESamp achieves robust generalization across mathematics, science, and code generation benchmarks and breaks the trade-off between diversity and coherence in creative writing. Our code has released at: https://github.com/LinesHogan/tLLM.

Users typically interact with and evaluate language models via single outputs, but each output is just one sample from a broad distribution of possible completions. This interaction hides distributional structure such as modes, uncommon edge cases, and sensitivity to small prompt changes, leading users to over-generalize from anecdotes when iterating on prompts for open-ended tasks. Informed by a formative study with researchers who use LMs (n=13) examining when stochasticity matters in practice, how they reason about distributions over language, and where current workflows break down, we introduce GROVE. GROVE is an interactive visualization that represents multiple LM generations as overlapping paths through a text graph, revealing shared structure, branching points, and clusters while preserving access to raw outputs. We evaluate across three crowdsourced user studies (N=47, 44, and 40 participants) targeting complementary distributional tasks. Our results support a hybrid workflow: graph summaries improve structural judgments such as assessing diversity, while direct output inspection remains stronger for detail-oriented questions.

Post-trained language models produce less varied outputs than their base counterparts. This output diversity collapse undermines inference-time scaling methods that rely on varied samples, and risks homogenizing model outputs on creative and value-laden tasks. Prior work attributes collapse to specific post-training methods, without separating the role of training data composition from the method, or the generation format from the model weights. We trace output diversity through three parallel post-training lineages of Olmo 3, Think (chain-of-thought distillation), Instruct (broad multi-source data), and RL-Zero, across 15 tasks and four text diversity metrics. We find that the location of collapse co-varies with data composition: the Think lineage loses most semantic diversity at supervised fine-tuning, and the effect of DPO is larger in Instruct than in Think. Suppressing chain-of-thought reasoning at inference in Think models drops accuracy on hard tasks, yet leaves answer-level diversity unchanged, showing that the collapse is embedded in the model weights by training data, not imposed by the generation format. Decomposing diversity loss on six verifiable tasks into a quality-control component (removal of incorrect outputs) and a residual component (genuine narrowing among correct outputs) reveals that the split is task-dependent, and Think models retain more correct-answer diversity than Instruct despite collapsing more in aggregate. Our results indicate that diversity collapse is determined during training by data composition and cannot be addressed at inference time alone.

Large language models produce repetitive output when prompted independently across many batches, a phenomenon we term cross-batch mode collapse: the progressive loss of output diversity when a language model is prompted repeatedly without access to its prior generations. Practitioners have long mitigated this with ad hoc deduplication and seed rotation, but no principled framework exists. We introduce Dynamic Context Evolution (DCE), comprising three mechanisms: (1) verbalized tail sampling (the model labels each idea with a guess about how obvious it is, and obvious ideas are discarded), which filters high-probability candidates via model self-assessment; (2) semantic memory, which maintains a persistent embedding index to reject near-duplicates across batches; and (3) adaptive prompt evolution, which reconstructs the generation prompt each batch using memory state and rotating diversity strategies. In experiments across three domains (sustainable packaging concepts, educational exam questions, and creative writing prompts) and two model families (gpt-5-mini and claude-haiku-4-5), a component ablation across 2-3 random seeds per method shows that DCE achieves 0.0 +/- 0.0% collapse versus 5.6 +/- 2.0% for naive prompting, while producing 17-18 HDBSCAN clusters per seed versus naive's volatile 2-17, indicating reliably richer conceptual structure. These results are validated with an independent embedding model (all-MiniLM-L6-v2) and hold across sensitivity sweeps of the VTS threshold tau and dedup threshold delta. Deduplication and prompt evolution are individually insufficient but jointly effective, at approximately $0.50 per 1,000 candidates using only standard API calls, with no fine-tuning or custom architectures required.

Research on Large Language Models (LLMs) studies output variation across generation, reasoning, alignment, and representational analysis, often under the umbrella of "diversity." Yet the terminology remains fragmented, largely because the normative objectives underlying tasks are rarely made explicit. We introduce the Magic, Madness, Heaven, Sin framework, which models output variation along a homogeneity-heterogeneity axis, where valuation is determined by the task and its normative objective. We organize tasks into four normative contexts: epistemic (factuality), interactional (user utility), societal (representation), and safety (robustness). For each, we examine the failure modes and vocabulary such as hallucination, mode collapse, bias, and erasure through which variation is studied. We apply the framework to analyze all pairwise cross-contextual interactions, revealing that optimizing for one objective, such as improving safety, can inadvertently harm demographic representation or creative diversity. We argue for context-aware evaluation of output variation, reframing it as a property shaped by task objectives rather than a model's intrinsic trait.

Diverse outputs in text generation are necessary for effective exploration in complex reasoning tasks, such as code generation and mathematical problem solving. Such Pass@$k$ problems benefit from distinct candidates covering the solution space. However, traditional sampling approaches often waste computational resources on repetitive failure modes. While Diffusion Language Models have emerged as a competitive alternative to the prevailing Autoregressive paradigm, they remain susceptible to this redundancy, with independent samples frequently collapsing into similar modes. To address this, we propose a training free, low cost intervention to enhance generative diversity in Diffusion Language Models. Our approach modifies intermediate samples in a batch sequentially, where each sample is repelled from the feature space of previous samples, actively penalising redundancy. Unlike prior methods that require retraining or beam search, our strategy incurs negligible computational overhead, while ensuring that each sample contributes a unique perspective to the batch. We evaluate our method on the HumanEval and GSM8K benchmarks using the LLaDA-8B-Instruct model. Our results demonstrate significantly improved diversity and Pass@$k$ performance across various temperature settings. As a simple modification to the sampling process, our method offers an immediate, low-cost improvement for current and future Diffusion Language Models in tasks that benefit from diverse solution search. We make our code available at https://github.com/sean-lamont/odd.

Aligning large language models (LLMs) to serve users with heterogeneous and potentially conflicting preferences is a central challenge for personalized and trustworthy AI. We formalize an ideal notion of universal alignment through test-time scaling: for each prompt, the model produces $k\ge 1$ candidate responses and a user selects their preferred one. We introduce $(k,f(k))$-robust alignment, which requires the $k$-output model to have win rate $f(k)$ against any other single-output model, and asymptotic universal alignment (U-alignment), which requires $f(k)\to 1$ as $k\to\infty$. Our main result characterizes the optimal convergence rate: there exists a family of single-output policies whose $k$-sample product policies achieve U-alignment at rate $f(k)=\frac{k}{k+1}$, and no method can achieve a faster rate in general. We show that popular post-training methods, including Nash learning from human feedback (NLHF), can fundamentally underutilize the benefits of test-time scaling. Even though NLHF is optimal for $k=1$, sampling from the resulting (often deterministic) policy cannot guarantee win rates above $\tfrac{1}{2}$ except for an arbitrarily small slack. This stems from a lack of output diversity: existing alignment methods can collapse to a single majority-preferred response, making additional samples redundant. In contrast, our approach preserves output diversity and achieves the optimal test-time scaling rate. In particular, we propose a family of symmetric multi-player alignment games and prove that any symmetric Nash equilibrium policy of the $(k+1)$-player alignment game achieves the optimal $(k,\frac{k}{k+1})$-robust alignment. Finally, we provide theoretical convergence guarantees for self-play learning dynamics in these games and extend the framework to opponents that also generate multiple responses.

Language models (LMs) often struggle to generate diverse, human-like creative content, raising concerns about the long-term homogenization of human thought through repeated exposure to similar outputs. Yet scalable methods for evaluating LM output diversity remain limited, especially beyond narrow tasks such as random number or name generation, or beyond repeated sampling from a single model. We introduce Infinity-Chat, a large-scale dataset of 26K diverse, real-world, open-ended user queries that admit a wide range of plausible answers with no single ground truth. We introduce the first comprehensive taxonomy for characterizing the full spectrum of open-ended prompts posed to LMs, comprising 6 top-level categories (e.g., brainstorm & ideation) that further breaks down to 17 subcategories. Using Infinity-Chat, we present a large-scale study of mode collapse in LMs, revealing a pronounced Artificial Hivemind effect in open-ended generation of LMs, characterized by (1) intra-model repetition, where a single model consistently generates similar responses, and more so (2) inter-model homogeneity, where different models produce strikingly similar outputs. Infinity-Chat also includes 31,250 human annotations, across absolute ratings and pairwise preferences, with 25 independent human annotations per example. This enables studying collective and individual-specific human preferences in response to open-ended queries. Our findings show that LMs, reward models, and LM judges are less well calibrated to human ratings on model generations that elicit differing idiosyncratic annotator preferences, despite maintaining comparable overall quality. Overall, INFINITY-CHAT presents the first large-scale resource for systematically studying real-world open-ended queries to LMs, revealing critical insights to guide future research for mitigating long-term AI safety risks posed by the Artificial Hivemind.

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.