Search: “safety”

Am old enough to remember when @GeoffreyHinton told me I was stupid for saying that LLMs regurgitate training data. He was wrong. LLM regurg…

Am old enough to remember when @GeoffreyHinton told me I was stupid for saying that LLMs regurgitate training data. He was wrong. LLM regurgitation is now one of the best-established findings in the f

Safety

Anisotropic Modality Align

DGX agent

arXiv:2605.07825v1 Announce Type: cross Abstract: Training multimodal large language models has long been limited by the scarcity of high-quality paired multimodal data. Recent studies show that the s

safetyarxiv-cs-cv

11 May 2026

APEX: Assumption-free Projection-based Embedding eXamination Metric for Image Quality Assessment

arXiv:2605.07786v1 Announce Type: cross Abstract: As generative models achieve unprecedented visual quality, the gold standard for image evaluation remains traditional feature-distribution metrics (e.

ART for Diffusion Sampling: A Reinforcement Learning Approach to Timestep Schedule

arXiv:2601.18681v2 Announce Type: replace-cross Abstract: We consider time discretization for score-based diffusion models to generate samples from a learned reverse-time dynamic on a finite grid. Uni

ASPECT: Node-Level Adaptive Spectral Fusion for Graph Contrastive Learning

arXiv:2604.01878v2 Announce Type: replace-cross Abstract: Spectral graph contrastive learning often constructs low- and high-frequency views to capture complementary graph signals, but these views are

Asymmetric On-Policy Distillation: Bridging Exploitation and Imitation at the Token Level

arXiv:2605.06387v2 Announce Type: replace-cross Abstract: On-policy distillation (OPD) trains a student on its own trajectories with token-level teacher feedback and often outperforms off-policy disti

Bellman Calibration for V-Learning in Offline Reinforcement Learning

arXiv:2512.23694v2 Announce Type: replace-cross Abstract: Reliable long-horizon value prediction is difficult in offline reinforcement learning because fitted value methods combine bootstrapping, func

Better Protein Function Prediction by Modeling Survivorship Bias

arXiv:2605.06879v1 Announce Type: new Abstract: Protein sequence data from nature exhibits survivorship bias: we only observe data from those organisms that survive and reproduce, while non-functional

Beyond Pairs: Your Language Model is Secretly Optimizing a Preference Graph

arXiv:2605.08037v1 Announce Type: cross Abstract: Direct Preference Optimization (DPO) aligns language models using pairwise preference comparisons, offering a simple and effective alternative to Rein

Beyond State-Wise Mirror Descent: Offline Policy Optimization with Parametric Policies

arXiv:2602.23811v4 Announce Type: replace-cross Abstract: We investigate the theoretical aspects of offline reinforcement learning (RL) under general function approximation. While prior works (e.g., X

Bias and Uncertainty in LLM-as-a-Judge Estimation

arXiv:2605.06939v1 Announce Type: new Abstract: LLM-as-a-Judge evaluation has become a standard tool for assessing base model performance. However, characterizing performance via the naive estimator,

🚨BREAKING: Ilya just confirmed under oath what he saw: Sam lying. And he confirmed that he thought it was appropriate to fire Altman for it…

🚨BREAKING: Ilya just confirmed under oath what he saw: Sam lying. And he confirmed that he thought it was appropriate to fire Altman for it. And that he had been concerned for about Sam for a long tim

CalexNet: Soft Cascade-Aligned Training and Calibration for Lightweight Early-Exit Branches

arXiv:2509.08318v2 Announce Type: replace Abstract: Early-exit cascades over a frozen convolutional backbone enable adaptive inference but suffer from three sources of train-inference mismatch: branch

Can David Beat Goliath? On Multi-Hop Reasoning with Resource-Constrained Agents

arXiv:2601.21699v2 Announce Type: replace Abstract: Multi-turn reasoning agents solve complex questions by decomposing them into intermediate retrieval or tool-use steps, for accumulating supporting e

safetyarxiv-cs-cl

Causal EpiNets: Precision-corrected Bounds on Individual Treatment Effects using Epistemic Neural Networks

arXiv:2605.07065v1 Announce Type: cross Abstract: Individual treatment effects are not point-identified from data. The Probability of Necessity and Sufficiency (PNS) circumvents this limitation by cha

Checkmate to everyone on X who doubted my coverage of Sam’s firing. Checkmate.

Checkmate to everyone on X who doubted my coverage of Sam’s firing. Checkmate. 🚨BREAKING: Ilya just confirmed under oath what he saw: Sam lying. And he confirmed that he thought it was appropriate to

Cognitive Agent Compilation for Explicit Problem Solver Modeling

arXiv:2605.07040v1 Announce Type: cross Abstract: Large language models (LLMs) are widely used for tutoring, feedback generation, and content creation, but their broad pretraining makes them hard to c

Compute is scarce and has to be bought years in advance, before companies know whether revenue will ever catch up. That is exactly the natur…

Compute is scarce and has to be bought years in advance, before companies know whether revenue will ever catch up. That is exactly the nature of the largest gamble in history. Heaven help the global e

Conditional generation of antibody sequences with classifier-guided germline-absorbing discrete diffusion

arXiv:2605.06720v1 Announce Type: cross Abstract: Antibody therapeutics are among the most successful modern medicines, yet computationally designing antibodies with desirable binding and developabili

Confidence-Aware Alignment Makes Reasoning LLMs More Reliable

arXiv:2605.07353v1 Announce Type: new Abstract: Large reasoning models often reach correct answers through flawed intermediate steps, creating a gap between final accuracy and reasoning reliability. E

Conformal-Style Quantile Analyses for Stochastic Bandits

arXiv:2605.07115v1 Announce Type: new Abstract: Stochastic bandit algorithms are usually analyzed under a mean-reward criterion, yet many problems favor arms with strong upper-tail performance, which

Contact-Grounded Policy: Dexterous Visuotactile Policy with Generative Contact Grounding

arXiv:2603.05687v3 Announce Type: replace Abstract: Contact-rich dexterous manipulation with multi-finger hands remains an open challenge in robotics because task success depends on multi-point contac

safetyarxiv-cs-ro

Cost-Ordered Feasibility for Multi-Armed Bandits with Cost Subsidy

arXiv:2605.07171v1 Announce Type: new Abstract: The classic multi-armed bandit (MAB) problem tackles the challenge of accruing maximum reward while making decisions under uncertainty. However, in appl

Curated Synthetic Data Doesn't Have to Collapse: A Theoretical Study of Generative Retraining with Pluralistic Preferences

arXiv:2605.07724v1 Announce Type: cross Abstract: Recursive retraining of generative models poses a critical representation challenge: when synthetic outputs are curated based on a fixed reward signal

DCGL: Dual-Channel Graph Learning with Large Language Models for Knowledge-Aware Recommendation

arXiv:2605.07314v1 Announce Type: cross Abstract: Knowledge Graphs (KGs) have proven highly effective for recommendation systems by capturing latent item relationships, while recent integration of Lar

Dear @geoffreyhinton, I literally never said that AI systems “JUST regurgitate”; that’s plainly false. I don’t believe it, and I didn’t say …

Dear @geoffreyhinton, I literally never said that AI systems “JUST regurgitate”; that’s plainly false. I don’t believe it, and I didn’t say it. (They do *sometimes* regurgitate, and the evidence for t

Decentralized Time-Varying Optimization for Streaming Data via Temporal Weighting

arXiv:2605.06971v1 Announce Type: cross Abstract: Classical optimization theory largely focuses on fixed objective functions, whereas many modern learning systems operate in dynamic environments where

Decoupling Semantics and Fingerprints: A Universal Representation for AI-Generated Image Detection

arXiv:2605.07074v1 Announce Type: new Abstract: Detecting AI-generated images across unseen architectures remains challenging, as existing models often overfit to generator-specific fingerprints and s

DiffeoMorph: Learning to Morph 3D Shapes Using Differentiable Agent-Based Simulations

arXiv:2512.17129v2 Announce Type: replace Abstract: Biological systems can form complex three-dimensional structures through the collective behavior of agents that share a common update rule and opera

Differentially Private Auditing Under Strategic Response

arXiv:2605.07674v1 Announce Type: cross Abstract: Regulatory audits of AI systems increasingly rely on differential privacy (DP) to protect training data and model internals. We study audit design whe

Diffusion-APO: Trajectory-Aware Direct Preference Alignment for Video Diffusion Transformers

arXiv:2605.07503v1 Announce Type: new Abstract: Efficiently aligning large-scale video diffusion models with human intent requires a scalable and trajectory-aware pathway that bridges the inherent dis

Direction-Flipped Influence Audits Reveal Hidden Structure in Moral Choices of LLMs

arXiv:2602.22831v2 Announce Type: replace-cross Abstract: Moral benchmarks for LLMs typically score models on context-free prompts, implicitly treating the measured choice rate as stable. We test this

Discovering Multiagent Learning Algorithms with Large Language Models

arXiv:2602.16928v3 Announce Type: replace-cross Abstract: Much of the advancement in Multi-Agent Reinforcement Learning (MARL) for imperfect-information games has historically depended on the manual,

Does Your Neural Network Extrapolate? Feature Engineering as Identifiability Bias for OOD Generalization

arXiv:2605.07483v1 Announce Type: cross Abstract: Successful deep neural networks discover salient features of data. We show when and why they fail to learn out-of-distribution (OOD)-relevant represen

Don't Retrain, Align: Adapting Autoregressive LMs to Diffusion LMs via Representation Alignment

arXiv:2605.06885v1 Announce Type: cross Abstract: Diffusion language models (DLMs) have recently demonstrated capabilities that complement standard autoregressive (AR) models, particularly in non-sequ

Dr-BA: Separable Optimization for Direct Radar Bundle Adjustment & Localization

arXiv:2605.07041v1 Announce Type: cross Abstract: This paper introduces Dr-BA, a first-of-its-kind radar bundle adjustment (BA) framework that operates directly on 2D spinning radar intensity images.

Dr. Post-Training: A Data Regularization Perspective on LLM Post-Training

arXiv:2605.07063v1 Announce Type: cross Abstract: Data selection methods address a critical challenge in LLM post-training: effectively leveraging scarce, high-fidelity target data alongside abundant

DReS: Dual Reconstruction Smoothing for Functional Regularization

arXiv:2510.00253v2 Announce Type: replace Abstract: Smoothness is a key inductive bias in machine learning and is closely related to generalization. Existing smoothness-inducing methods typically rely

Drifting Field Policy: A One-Step Generative Policy via Wasserstein Gradient Flow

arXiv:2605.07727v1 Announce Type: cross Abstract: We propose Drifting Field Policy (DFP), a non-ODE one-step generative policy built on the drifting model paradigm. We frame the policy update as a rev

Dynamic one-time delivery of critical data by small and sparse UAV swarms: a model problem for MARL scaling studies

arXiv:2512.09682v2 Announce Type: replace-cross Abstract: This work studies the application of Multi-Agent Reinforcement Learning (MARL) to decentralized control of unmanned aerial vehicles to relay a

Econometric vs. Causal Structure-Learning for Time-Series Policy Decisions: Evidence from the UK COVID-19 Policies

arXiv:2603.00041v2 Announce Type: replace-cross Abstract: Causal machine learning (ML) recovers graphical structures that inform us about potential cause-and-effect relationships. Most progress has fo

EditTransfer++: Toward Faithful and Efficient Visual-Prompt-Guided Image Editing

arXiv:2605.07455v1 Announce Type: new Abstract: Visual-prompt-guided edit transfer aims to learn image transformations directly from example pairs, offering more precise and controllable editing than

EGA: Adapting Frozen Encoders for Vector Search with Bounded Out-of-Distribution Degradation

arXiv:2605.05674v2 Announce Type: replace-cross Abstract: Vector search systems built on frozen vision encoders face queries from unseen classes at deployment, yet existing adapter training collapses

Emergent social transmission of model-based representations without inference

arXiv:2604.05777v2 Announce Type: replace Abstract: How do people acquire rich, flexible knowledge about their environment from others despite limited cognitive capacity? Humans are often thought to r

Emergent Symbolic Structure in Health Foundation Models: Extraction, Alignment, and Cross-Modal Transfer

arXiv:2605.07407v1 Announce Type: new Abstract: Health foundation models (FMs) learn useful representations from wearable sensors, but interpreting what they encode and transferring that knowledge acr

Entropy-Regularized Adjoint Matching for Offline Reinforcement Learning

arXiv:2605.06156v2 Announce Type: replace-cross Abstract: Integrating expressive generative policies, such as flow-matching models, into offline reinforcement learning (RL) allows agents to capture co

Epistemological standards in SF venture capital are so low that you can write this, and everyone will simply nod along. The actual evidence,…

Epistemological standards in SF venture capital are so low that you can write this, and everyone will simply nod along. The actual evidence, including a recent NBER paper on thousands of firms, finds:

Escaping the Diversity Trap in Robotic Manipulation via Anchor-Centric Adaptation

arXiv:2605.07381v1 Announce Type: cross Abstract: While Vision-Language-Action (VLA) models offer broad general capabilities, deploying them on specific hardware requires real-world adaptation to brid