The research we build on.

• 2026-02arXiv 2602.19041
  Back to Blackwell: Closing the Loop on Intransitivity in Multi-Objective Preference Fine-Tuning

  Jiahao Zhang, Lujing Zhang, Keltin Grimes et al. · Machine learning

  , cyclic) preferences. Intransitive preferences often stem from either $\textit{(i)}$ inconsistent rankings along a single objective or $\textit{(ii)}$ scalarizing multiple objectives into a single metric.

  Cited inStudy: Logical coherence of LLM world models
• 2026-02arXiv 2603.06612
  Consensus is Not Verification: Why Crowd Wisdom Strategies Fail for LLM Truthfulness

  Yegor Denisov-Blanch, Joshua Kazdan, Jessica Chudnovsky et al. · Machine learning

  Pass@k and other methods of scaling inference compute can improve language model performance in domains with external verifiers, including mathematics and code, where incorrect candidates can be filtered reliably.

  Cited inStudy: Orchestrated specialists and bridge synthesisStudy: Stopping confident wrong expert votes
• 2026-02arXiv 2603.06616
  RACER: Risk-Aware Calibrated Efficient Routing for Large Language Models

  Sai Hao, Hao Zeng, Hongxin Wei et al. · Machine learning

  Efficiently routing queries to the optimal large language model (LLM) is crucial for optimizing the cost-performance trade-off in multi-model systems. However, most existing routers rely on single-model selection, making them susceptible to misrouting.

  Cited inRoute on shared statistics, never expert confidence: the router architecture for frozen expertsStudy: Optimal router architecture
• 2026-02arXiv 2602.17053
  RFEval: Benchmarking Reasoning Faithfulness under Counterfactual Reasoning Intervention in Large Reasoning Models

  Yunseok Han, Yejoon Lee, Jaeyoung Do · AI & agents

  Large Reasoning Models (LRMs) exhibit strong performance, yet often produce rationales that sound plausible but fail to reflect their true decision process, undermining reliability and trust.

  Cited inStudy: Logical coherence of LLM world models
• 2026-02arXiv 2602.17559
  Revisiting Weight Regularization for Low-Rank Continual Learning

  Yaoyue Zheng, Yin Zhang, Joost van de Weijer et al. · Machine learning

  Continual Learning (CL) with large-scale pre-trained models (PTMs) has recently gained wide attention, shifting the focus from training from scratch to continually adapting PTMs.

  Cited inEWC protects what its Fisher sees: a literature synthesis on elastic weight consolidation, scale and coherenceStudy: Elastic weight consolidation deep dive
• 2026-02arXiv 2602.16653
  Agent Skill Framework: Perspectives on the Potential of Small to Medium Language Models in Industrial Environments

  Yangjie Xu, Lujun Li, Lama Sleem et al. · AI & agents

  Skills are getting increasingly used in recent agentic frameworks, performing robustly with proprietary LLMs. Yet, their effectiveness for small and medium-sized open-source models (270M--80B) remains underexplored, especially under industrial, resource-constrained conditions.

  Cited inStudy: Forced vs routed skill selectionStudy: Genesys program prior art
• 2026-02arXiv 2602.16931
  Narrow Fine-Tuning Erodes Safety Alignment in Vision-Language Agents

  Idhant Gulati, Shivam Raval · AI & agents

  Lifelong multimodal agents must continuously adapt to new tasks through post-training, but this creates a fundamental tension between acquiring capabilities and preserving safety alignment.

  Cited inStudy: LLMs modifying open model weights
• 2026-02arXiv 2602.14111
  Sanity Checks for Sparse Autoencoders: Do SAEs Beat Random Baselines?

  Anton Korznikov, Andrey Galichin, Alexey Dontsov et al. · Machine learning

  Sparse Autoencoders (SAEs) have emerged as a promising tool for interpreting neural networks by decomposing their activations into sparse sets of human-interpretable features.

  Cited inStudy: Decision signatures for conflict detection
• 2026-02arXiv 2602.12587
  Multi-Head Attention as a Source of Catastrophic Forgetting in MoE Transformers

  Anrui Chen, Ruijun Huang, Xin Zhang et al. · Machine learning

  Mixture-of-Experts (MoE) architectures are often considered a natural fit for continual learning because sparse routing should localize updates and reduce interference, yet MoE Transformers still forget substantially even with sparse, well-balanced expert utilization.

  Cited inStudy: Merging and splitting MoE experts
• 2026-02arXiv 2602.12634
  The Rise of AI Agent Communities: Large-Scale Analysis of Discourse and Interaction on Moltbook

  Lingyao Li, Renkai Ma, Chen Chen et al. · Society, software & economics

  Moltbook is a Reddit-like social platform where AI agents create posts and interact with other agents through comments and replies, offering a real-world setting to examine agent-to-agent communication at scale.

  Cited inStudy: Civilization as a superintelligence model
• 2026-02arXiv 2602.13093
  Consistency of Large Reasoning Models Under Multi-Turn Attacks

  Yubo Li, Ramayya Krishnan, Rema Padman · AI & agents

  Large reasoning models with reasoning capabilities achieve state-of-the-art performance on complex tasks, but their robustness under multi-turn adversarial pressure remains underexplored.

  Cited inStable wrong beliefs evade every self-check: a literature synthesis on model self-knowledge and confident errorsStudy: Model self knowledge and confident errors
• 2026-02arXiv 2602.13458
  MoltNet: Understanding Social Behavior of AI Agents in the Agent-Native MoltBook

  Yi Feng, Chen Huang, Zhibo Man et al. · Society, software & economics

  Large-scale communities of AI agents are becoming increasingly prevalent, creating new environments for agent-agent social interaction. Prior work has examined multi-agent behavior primarily in controlled or small-scale settings, limiting our understanding of emergent social dynamics at scale.

  Cited inStudy: Civilization as a superintelligence model
• 2026-02arXiv 2602.15064
  Structural Divergence Between AI-Agent and Human Social Networks in Moltbook

  Wenpin Hou, Zhicheng Ji · Hardware, photonics & physics

  Large populations of AI agents are increasingly embedded in online environments, yet little is known about how their collective interaction patterns compare to human social systems.

  Cited inStudy: Civilization as a superintelligence model
• 2026-02arXiv 2602.15902
  Doc-to-LoRA: Learning to Instantly Internalize Contexts

  Rujikorn Charakorn, Edoardo Cetin, Shinnosuke Uesaka et al. · Language models & NLP

  Long input sequences are central to in-context learning, document understanding, and multi-step reasoning of Large Language Models (LLMs). However, the quadratic attention cost of Transformers makes inference memory-intensive and slow.

  Cited inA model rewrote another model's weights. Read what it took away.Study: LLMs modifying open model weights
• 2026-02arXiv 2603.09988
  Causally Grounded Mechanistic Interpretability for LLMs with Faithful Natural-Language Explanations

  Ajay Pravin Mahale · Language models & NLP

  Mechanistic interpretability identifies internal circuits responsible for model behaviors, yet translating these findings into human-understandable explanations remains an open problem.

  Cited inStudy: Weight editing to prevent forgetting
• 2026-02arXiv 2602.11686
  LAER-MoE: Load-Adaptive Expert Re-layout for Efficient Mixture-of-Experts Training

  Xinyi Liu, Yujie Wang, Fangcheng Fu et al. · Other

  Expert parallelism is vital for effectively training Mixture-of-Experts (MoE) models, enabling different devices to host distinct experts, with each device processing different input data.

  Cited inStudy: Merging and splitting MoE experts
• 2026-02arXiv 2602.11137
  Weight Decay Improves Language Model Plasticity

  Tessa Han, Sebastian Bordt, Hanlin Zhang et al. · Machine learning

  Large language models are typically trained in two broad phases: pretraining to produce a base model, followed by further training to improve downstream performance.

  Cited inStudy: Fortifying knowledge in continual learning
• 2026-02arXiv 2602.09316
  Effective MoE-based LLM Compression by Exploiting Heterogeneous Inter-Group Experts Routing Frequency and Information Density

  Zhendong Mi, Yixiao Chen, Pu Zhao et al. · Machine learning

  Mixture-of-Experts (MoE) based Large Language Models (LLMs) have achieved superior performance, yet the massive memory overhead caused by storing multiple expert networks severely hinders their practical deployment.

  Cited inStudy: Merging and splitting MoE experts
• 2026-02arXiv 2602.09870
  Steer2Edit: From Activation Steering to Component-Level Editing

  Chung-En Sun, Ge Yan, Zimo Wang et al. · Language models & NLP

  Steering methods influence Large Language Model behavior by identifying semantic directions in hidden representations, but are typically realized through inference-time activation interventions that apply a fixed, global modification to the model's internal states.

  Cited inStudy: Weight editing to prevent forgetting
• 2026-02arXiv 2602.08218
  Sparsity-Aware Evolution for Model Merging

  Huan Zhang, Yanjian Zhang, Guillaume Wisniewski et al. · Machine learning

  We propose a sparsity-aware evolutionary (SAE) framework for model merging that involves iterative pruning-merging cycles to act as a novel mutation operator.

  Cited inStudy: Weight editing to prevent forgetting
• 2026-02arXiv 2602.08040
  FIRE: Frobenius-Isometry Reinitialization for Balancing the Stability-Plasticity Tradeoff

  Isaac Han, Sangyeon Park, Seungwon Oh et al. · Machine learning

  , adapting to new tasks). Standard reinitialization methods, which reinitialize weights toward their original values, are widely used but difficult to tune: conservative reinitializations fail to restore plasticity, while aggressive ones erase useful knowledge.

  Cited inStudy: Fortifying knowledge in continual learning
• 2026-02arXiv 2602.06358
  SHINE: A Scalable In-Context Hypernetwork for Mapping Context to LoRA in a Single Pass

  Yewei Liu, Xiyuan Wang, Yansheng Mao et al. · Language models & NLP

  We propose SHINE (Scalable Hyper In-context NEtwork), a scalable hypernetwork that can map diverse meaningful contexts into high-quality LoRA adapters for large language models (LLMs).

  Cited inStudy: LLMs modifying open model weights
• 2026-02arXiv 2602.06359
  Training Data Selection with Gradient Orthogonality for Efficient Domain Adaptation

  Xiyang Zhang, Yuanhe Tian, Hongzhi Wang et al. · Machine learning

  Fine-tuning large language models (LLMs) for specialized domains often necessitates a trade-off between acquiring domain expertise and retaining general reasoning capabilities, a phenomenon known as catastrophic forgetting.

  Cited inStudy: Weight editing to prevent forgetting
• 2026-02arXiv 2602.06665
  Not All Layers Need Tuning: Selective Layer Restoration Recovers Diversity

  Bowen Zhang, Meiyi Wang, Harold Soh · Language models & NLP

  Post-training improves instruction-following and helpfulness of large language models (LLMs) but often reduces generation diversity, which leads to repetitive outputs in open-ended settings, a phenomenon known as mode collapse.

  Cited inStudy: Weight editing to prevent forgetting
• 2026-02arXiv 2602.06923
  From Kepler to Newton: Inductive Biases Guide Learned World Models in Transformers

  Ziming Liu, Sophia Sanborn, Surya Ganguli et al. · Machine learning

  Can general-purpose AI architectures go beyond prediction to discover the physical laws governing the universe? True intelligence relies on "world models" -- causal abstractions that allow an agent to not only predict future states but understand the underlying governing dynamics.

  Cited inCoherence is selected, anchored, then pruned: a literature synthesis on logical coherence in LLM world modelsWhat a model lets into long-term memory matters more than how it learnsStudy: Logical coherence of LLM world modelsStudy: Memory world models and source trust
• 2026-02arXiv 2602.05370
  Mining or Synthesis? Rethinking Exploration Efficiency in Iterative Alignment of Mathematical Reasoning

  Jun Rao, Zixiong Yu, Xuebo Liu et al. · Language models & NLP

  Iterative Direct Preference Optimization (DPO) has emerged as a widely used paradigm for aligning Large Language Models on reasoning tasks. Existing approaches typically rely on Best-of-N sampling ($N\geq8$) to mine positive trajectories from the distribution tail.

  Cited inStudy: Genesys program prior art
• 2026-02arXiv 2602.13273
  MergePipe: A Budget-Aware Parameter Management System for Scalable LLM Merging

  Yuanyi Wang, Yanggan Gu, Zihao Wang et al. · Retrieval & data

  Large language model (LLM) merging has become a key technique in modern LLM development pipelines, enabling the integration of multiple task- or domain-specific expert models without retraining.

  Cited inStudy: LLMs modifying open model weights
• 2026-02arXiv 2602.04509
  Model-Dowser: Data-Free Importance Probing to Mitigate Catastrophic Forgetting in Multimodal Large Language Models

  Hyeontaek Hwang, Nguyen Dinh Son, Daeyoung Kim · Language models & NLP

  Fine-tuning Multimodal Large Language Models (MLLMs) on task-specific data is an effective way to improve performance on downstream applications. However, such adaptation often leads to a degradation in generalization on pretrained tasks, a phenomenon known as Catastrophic Forgetting.

  Cited inStudy: Weight editing to prevent forgetting
• 2026-02arXiv 2602.04521
  $C$-$ΔΘ$: Circuit-Restricted Weight Arithmetic for Selective Refusal

  Aditya Kasliwal, Pratinav Seth, Vinay Kumar Sankarapu · Language models & NLP

  Modern deployments require LLMs to enforce safety policies at scale, yet many controls rely on inference-time interventions that add recurring compute cost and serving complexity.

  Cited inStudy: Weight editing to prevent forgetting
• 2026-02arXiv 2602.03473
  Scaling Continual Learning to 300+ Tasks with Bi-Level Routing Mixture-of-Experts

  Meng Lou, Yunxiang Fu, Yizhou Yu · Machine learning

  Continual learning, especially class-incremental learning (CIL), on the basis of a pre-trained model (PTM) has garnered substantial research interest in recent years.

  Cited inRoute on shared statistics, never expert confidence: the router architecture for frozen expertsStudy: Merging and splitting MoE expertsStudy: Optimal router architecture
• 2026-02arXiv 2602.03541
  Group Selection as a Safeguard Against AI Substitution

  Qiankun Zhong, Thomas F. Eisenmann, Julian Garcia et al. · AI & agents

  Reliance on generative AI can reduce cultural variance and diversity, especially in creative work. This reduction in variance has already led to problems in model performance, including model collapse and hallucination.

  Cited inStudy: Civilization as a superintelligence model
• 2026-02arXiv 2602.03747
  LIVE: Long-horizon Interactive Video World Modeling

  Junchao Huang, Ziyang Ye, Xinting Hu et al. · Vision

  Autoregressive video world models predict future visual observations conditioned on actions. While effective over short horizons, these models often struggle with long-horizon generation, as small prediction errors accumulate over time.

  Cited inStudy: Memory world models and source trust
• 2026-02arXiv 2602.10131
  The Anatomy of the Moltbook Social Graph

  David Holtz · Society, software & economics

  5 days (6{,}159 agents; 13{,}875 posts; 115{,}031 comments).91$). At the micro level, patterns appear distinctly non-human.1\% of messages are exact duplicates of viral templates.

  Cited inStudy: Civilization as a superintelligence model
• 2026-02arXiv 2602.01990
  SAME: Stabilized Mixture-of-Experts for Multimodal Continual Instruction Tuning

  Zhen-Hao Xie, Jun-Tao Tang, Yu-Cheng Shi et al. · Machine learning

  Multimodal Large Language Models (MLLMs) achieve strong performance through instruction tuning, but real-world deployment requires them to continually expand their capabilities, making Multimodal Continual Instruction Tuning (MCIT) essential.

  Cited inInternal notes
• 2026-02arXiv 2602.02132
  There Is More to Refusal in Large Language Models than a Single Direction

  Faaiz Joad, Majd Hawasly, Sabri Boughorbel et al. · Language models & NLP

  Prior work argues that refusal in large language models is mediated by a single direction, enabling steering and abliteration.

  Cited inStudy: Decision signatures for conflict detection
• 2026-02arXiv 2602.00943
  Dynamic Prior Thompson Sampling for Cold-Start Exploration in Recommender Systems

  Zhenyu Zhao, David Zhang, Ellie Zhao et al. · Machine learning

  Cold-start exploration is a core challenge in large-scale recommender systems: new or data-sparse items must receive traffic to estimate value, but over-exploration harms users and wastes impressions.

  Cited inStudy: Generational competitive continual learning
• 2026-02arXiv 2602.01270
  Mixture-of-World Models: Scaling Multi-Task Reinforcement Learning with Modular Latent Dynamics

  Boxuan Zhang, Weipu Zhang, Zhaohan Feng et al. · Machine learning

  A fundamental challenge in multi-task reinforcement learning (MTRL) is achieving sample efficiency in visual domains where tasks exhibit substantial heterogeneity in both observations and dynamics.

  Cited inStudy: Genesys program prior art
• 2026-01arXiv 2602.02568
  Mitigating Task-Order Sensitivity and Forgetting via Hierarchical Second-Order Consolidation

  Protik Nag, Krishnan Raghavan, Vignesh Narayanan · Machine learning

  We introduce $\textbf{Hierarchical Taylor Series-based Continual Learning (HTCL)}$, a framework that couples fast local adaptation with conservative, second-order global consolidation to address the high variance introduced by random task ordering.

  Cited inStudy: Continual learning order and companion data
• 2026-01arXiv 2602.02543
  Norm Anchors Make Model Edits Last

  Mingda Liu, Zhenghan Zhu, Ze'an Miao et al. · Machine learning

  Sequential Locate-and-Edit (L&E) model editing can fail abruptly after many edits.

  Cited inStudy: Consequence consistent updatesStudy: Continual learning order and companion dataStudy: Hardware guardian core for continual learningStudy: Seven continual learning methods tested
• 2026-01arXiv 2601.21183
  Sycophantic Anchors: Localizing and Quantifying User Agreement in Reasoning Models

  Jacek Duszenko · AI & agents

  Reasoning models frequently agree with incorrect user suggestions -- a behavior known as sycophancy. However, it is unclear where in the reasoning trace this agreement originates and how strong the commitment is.

  Cited inStudy: Decision signatures for conflict detection

Built from every arXiv paper cited in our papers, posts, topic hubs, reports and research notes, with titles, authors, dates and abstracts from the arXiv API (updated 2026-11-12). Summaries are the opening sentences of each paper's own abstract, not our assessment; a citation means we used or checked the work, not that we endorse it. Fields follow each paper's primary arXiv category. Papers cited only outside arXiv (journals, proceedings) are not listed yet. Entries are generated from our citations, so a mistyped ID in a note would list the wrong paper; we remove those as we find them. Report a wrong entry.