1
Scaling Self-Play with Self-Guidance
自博弈遇上自指导：新方法以自我引导驱动大规模训练，显著提升智能体策略多样性。
arXiv:2604.20209v2 Announce Type: replace Abstract: LLM self-play algorithms are notable in that, in principle, nothing bounds their learning: a Conje…
2
On the Effect of Sampling Diversity in Scaling LLM Inference
采样多样性如何左右大模型推理扩展的缩放规律？这项研究给出了新视角。
arXiv:2502.11027v5 Announce Type: replace Abstract: Large language model (LLM) scaling inference is key to unlocking greater performance, and leveragi…
3
On the Diversity of Analogy Making in Large Language Models
大模型类比能力多样性研究揭示AI推理新维度，或为认知智能评估提供全新视角。
arXiv:2608.03233v1 Announce Type: new Abstract: Large Language Models (LLMs) have demonstrated remarkable potential for analogy making, a core cogniti…
4
BODHI: Do LLMs Branch Out and Discover Heterogeneous Inferences?
探究大模型能否跳出单一推理路径，自主发现多元异构推断，视角新颖。
arXiv:2608.02867v1 Announce Type: new Abstract: Although reinforcement learning with verifiable rewards (RLVR) has improved the performance of large l…
5
Humans Are More Diverse: Frontier LLMs Show Extreme Policies in Idealised AI Development Races
前沿大模型在理想化AI竞赛中展露极端策略，而人类决策更趋多元，视角新颖的实验揭示两者差异。
arXiv:2608.01193v1 Announce Type: new Abstract: An AI development race creates a multi-agent safety dilemma. Each company can develop slowly and safel…
6
Why Silicon Valley's 'Neurodiversity Hiring' Programs Are Failing Autistic Workers (And How LLMs Are Quietly Fixing the Interview)
硅谷神经多样性招聘为何失灵？LLM正在悄然改写面试规则，让自闭症人才不再被埋没。
Why Silicon Valley's 'Neurodiversity Hiring' Programs Are Failing Autistic Workers (And How LLMs Are Quietly Fixing the Interview) Introduction: The G…
7
Collective Intelligence: The Next Frontier of AI
76,636个AI模型实时协作、辩论与批判，揭秘集体智能如何超越个体能力边界
Article URL: https://github.com/ailinone/collective-intelligence Comments URL: https://news.ycombinator.com/item?id=49053465 Points: 7 # Comments: 1
8
Are Diversity Metrics Measuring Diversity? A Capability-Controlled Audit of Majority-Vote Gain in LLM Ensembles
多数投票假设被质疑：这篇研究通过能力控制的审计方法，检验现有多样性指标是否真实衡量了LLM集成中的多样性，结果或有颠覆性。
arXiv:2607.20768v1 Announce Type: cross Abstract: Majority voting over LLMs is widely assumed to benefit from diversity, and diversity measures are us…
9
From Noise to Diversity: Random Embedding Injection in LLM Reasoning
注入随机噪声增强LLM推理多样性，ICML 2026 Spotlight论文新思路
arXiv:2605.11936v2 Announce Type: replace Abstract: Recent soft prompt research has tried to improve reasoning by inserting trained vectors into LLM i…
10
More Is Not More: What Matters for Diversity in LLM Opinions?
大模型模拟意见时，多样性并不靠堆数量，这项研究揭示了真正影响LLM意见多样性的关键因素
arXiv:2607.20429v1 Announce Type: cross Abstract: Large language models are increasingly used to simulate diverse human opinions in open-ended tasks s…
11
When Reasoning Narrows the Move: Diversity Collapse in LLM Game Play
LLM推理会扼杀创造力？这篇论文揭示AI在游戏中因过度推理导致策略多样性急剧下降的现象。
arXiv:2607.19523v1 Announce Type: new Abstract: Supervised fine-tuning (SFT) is widely used to adapt large language models to downstream tasks, but it…
12
Verbalized Sampling: How to Mitigate Mode Collapse and Unlock LLM Diversity
揭示LLM多样性丧失的深层数据原因，提出Verbalized Sampling方法缓解模式坍塌
arXiv:2510.01171v4 Announce Type: replace Abstract: Post-training alignment often reduces LLM diversity, leading to a phenomenon known as mode collaps…
13
Meta-Learning Preferences for Multilingual LLM Alignment
元学习攻克多语言LLM对齐难题，让偏好学习适应不同语言文化
arXiv:2607.13315v1 Announce Type: new Abstract: Unequal availability of human preference data across languages poses a significant challenge for align…
14
The Hitchhiker's Guide to Monoculture
Gordon Burtch这本指南直击AI与代码同质化痛点，提供破解单一种植的实用思路。
arXiv:2607.13077v1 Announce Type: cross Abstract: Large language models (LLMs) often produce homogeneous outputs, raising concerns that AI coding assi…
15
Thinking Seeds: Leveraging Historical Diversity for Position-Aware RL in LLMs
提出Thinking Seeds方法，利用历史多样性与位置感知强化学习，显著提升LLM推理效率与泛化能力。
arXiv:2601.21476v2 Announce Type: replace Abstract: On-policy reinforcement learning (RL) for language model post-training suffers from a fundamental …
16
LEMUR 2: Unlocking Neural Network Diversity for AI
从模型多样性切入，LEMUR 2 为AI泛化能力提供新思路，值得算法研究者细读。
arXiv:2607.06839v1 Announce Type: new Abstract: Existing NAS benchmarks (e.g., NAS-Bench, NATS-Bench) cover only narrow, task-specific regions of the …
17
Effectiveness of LLM-based Software Diversity for Reliability Improvement -- an Empirical Study
用大模型生成多样代码提升软件可靠性？这项实证研究给出严谨验证，值得开发者关注。
arXiv:2607.03174v1 Announce Type: cross Abstract: Software diversity has been extensively studied as a means of reducing the risk of common-mode failu…
18
EvoXplain: When Machine Learning Models Agree on Predictions but Disagree on Why -- Measuring Mechanistic Multiplicity Across Training Runs
预测相同、解释却大相径庭，这项研究为可解释机器学习敲响警钟。
arXiv:2512.22240v5 Announce Type: replace-cross Abstract: Machine learning models are primarily judged by predictive performance, especially in applie…
19
Simulating Tradeoffs in AI Organisation
用LLM模拟AI组织中的单极vs多极、隐藏vs可见代理等权衡，一个值得探索的新方向
Currently there doesn't seem to be anyone trying to simulate the tradeoffs between a singleton and a multipolar setup. Or looking at other tradeoffs l…
20
Inference-Time Diversity in RL-Trained Lean Theorem Provers: A Diagnostic Study
诊断强化学习训练的Lean定理证明器在推理时的多样性缺陷，为AI数学推理的鲁棒性提供新视角。
arXiv:2601.16172v3 Announce Type: replace Abstract: RL-trained Lean theorem provers mode-collapse at inference time: on miniF2F-test with DeepSeek-Pro…