Search: “safety”

Belief-State Engine: Augmenting LLMs for Principled Planning Under Partial Observability

arXiv:2609.10036v1 Announce Type: cross Abstract: Large language model agents produce fluent action sequences across a wide range of tasks, yet they fail in characteristic ways once the environment be

Content type

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

BETA-Labeling for Multilingual Dataset Construction in Low-Resource IR

DGX agent

arXiv:2602.14488v3 Announce Type: replace-cross Abstract: IR in low-resource languages remains limited by the scarcity of high-quality, task-specific annotated datasets. Manual annotation is expensive

safetyarxiv-cs-ai

10 Sept 2026

Beyond Retrieval: Joint Supervision and Multimodal Document Ranking for Textbook Question Answering

arXiv:2505.13520v2 Announce Type: replace-cross Abstract: Textbook question answering (TQA) is a complex task, requiring the interpretation of complex multimodal context. Although recent advances have

Beyond Right and Wrong: Evaluating Second-order Social Reasoning in Large Language Models

arXiv:2609.05437v1 Announce Type: new Abstract: Previous AI alignment efforts have focused primarily on first-order social norms -- teaching models what is socially acceptable or unacceptable (e.g., `

BRACE: Anchored Bellman-Residual Correction for Stale Critics in Asynchronous RL

arXiv:2609.09783v1 Announce Type: new Abstract: Asynchronous reinforcement learning has become the standard way to scale training for language models, but the resulting policy lag biases the critic to

Bridging the Semantic-Utility Gap in Multimodal RAG via Generator-in-the-Loop Alignment

arXiv:2609.08188v1 Announce Type: new Abstract: Vision-language models (VLMs) augmented with retrieval-augmented generation (RAG) benefit from access to external evidence. However, standard retrievers

BTBR: A Bayesian-Theory-Driven Probabilistic-Fuzzy Framework for Implicit Bias Removal in Large Language Models

arXiv:2408.10608v2 Announce Type: replace Abstract: Large language models (LLMs) may encode biased associations from heterogeneous training corpora that are not immediately visible under ordinary prom

CoGe-GCD: Reframing Generalized Category Discovery with Compositional Generalization

arXiv:2609.10158v1 Announce Type: new Abstract: Generalized Category Discovery (GCD) assigns unlabeled instances, mixed with labeled data, to known or novel categories, requiring human-like compositio

common belief, for sure! but also, as i discuss in my (free) newsletter today, ridiculous, do you @peterwildeford actually believe it? see a…

common belief, for sure! but also, as i discuss in my (free) newsletter today, ridiculous, do you @peterwildeford actually believe it? see any holes in my argument! OpenAI employee says there's 70% ch

Compact Visuotactile World Models for Lifting: Prediction, Reward Alignment, and Force Constraints

arXiv:2609.09597v1 Announce Type: new Abstract: Accurate contact prediction is useful for robotic manipulation only if it supports effective decisions. We investigate this connection using a compact,

CompassOPD: Cross-Family On-Policy Distillation via Within-Family Likelihood Shifts

arXiv:2609.10154v1 Announce Type: new Abstract: On-policy distillation (OPD) provides dense token-level supervision on student-generated trajectories. Although OPD performs strongly when teacher and s

Control Copy-Paste: Controllable Diffusion-Based Augmentation Method for Remote Sensing Few-Shot Object Detection

arXiv:2507.21816v1 Announce Type: cross Abstract: Few-shot object detection (FSOD) for optical remote sensing images aims to detect rare objects with only a few annotated bounding boxes. The limited t

Cost-Aware Post-Hoc Deferral Under Calibration and Shift: An Environmental AI Case Study

arXiv:2609.09235v1 Announce Type: cross Abstract: Choosing a deferral policy for a frozen classifier requires more than ranking uncertain cases: confidence may be miscalibrated, errors have unequal co

CTC-TTS: LLM-based dual-streaming text-to-speech with CTC alignment

arXiv:2602.19574v2 Announce Type: replace-cross Abstract: Large-language-model (LLM)-based text-to-speech (TTS) systems can generate natural speech, but most are not designed for low-latency dual-stre

DART: A DAG-Based Reputation and Incentive Framework via Blockchain-Enabled Governance for Trustworthy LLM Multi-Agent Collaboration

arXiv:2609.05529v1 Announce Type: cross Abstract: Large language model (LLM)-based multi-agent systems (MAS) predominantly rely on centralized orchestration and lack formal verification mechanisms for

Decoupled Self-Forcing Distillation for Streaming Talking Head Generation

arXiv:2609.10317v1 Announce Type: new Abstract: Streaming talking-head generation produces each frame as its driving audio arrives, yet fidelity and efficiency have so far pulled in opposite direction

Deep and shallow biases in language models

arXiv:2609.09901v1 Announce Type: new Abstract: Large language models often repeatedly select the same answer even when many alternatives are plausible. Prior work treats this concentration as bias, b

Deterministic Prompting for Speaker-Stable Low-Resource Greek TTS

arXiv:2609.10022v1 Announce Type: cross Abstract: Modern TTS systems approach human quality for high-resource languages but degrade when clean speech data is scarce. Modern Greek exemplifies this, lac

Differentially Private Average Treatment Effect Estimation by Propensity Score Blocking

arXiv:2609.09536v1 Announce Type: cross Abstract: Average treatment effect (ATE) estimation in observational studies is a fundamental statistical tool used frequently in social science, medicine, and

Efficient Fairness Auditing Across Guidance Scales in Text-to-Image Diffusion Models via Causal Abstraction

arXiv:2609.09486v1 Announce Type: cross Abstract: Fairness auditing of text-to-image diffusion models often requires generating large numbers of images across sampling configurations, making comprehen

Evaluating Model Retraining under Drift: Paired Comparisons of Cumulative Subgroup Disparity

arXiv:2609.09788v1 Announce Type: new Abstract: Choosing when to retrain a deployed classifier requires assessing subgroup error rates across the sequence of models used, including periods between upd

“Everyone killed” still seems to me to be an unrealistic exaggeration; I still don’t see a plausible path to that.

“Everyone killed” still seems to me to be an unrealistic exaggeration; I still don’t see a plausible path to that. OpenAI and Anthropic researchers agree: their approach to AI development could get ev

Everything in Moderation: Per-Domain Coverage Optima and Alignment-Resistant Domain Gaps in Multi-Domain Mid-Training

arXiv:2609.09081v1 Announce Type: new Abstract: Mid-training, the stage between pre-training and alignment, is where a model's per-domain data composition is typically set by data availability rather

Evidence-Aligned Entity Verification for Hallucination Detection in Retrieval-Augmented Generation

arXiv:2609.08267v1 Announce Type: new Abstract: Hallucination detection is crucial for large language models (LLMs), as hallucinated content creates significant barriers in applications requiring fact

False positive bias in AI-powered speech-based cognitive screening for multilingual English speakers in the UK

arXiv:2602.13047v2 Announce Type: replace Abstract: Conversational speech reveals early signs of cognitive decline, including dementia and mild cognitive impairment (MCI). AI models show promise for s

FastE: Readout-Triggered Token Compression for LLM Embedding Inference

arXiv:2609.08407v1 Announce Type: new Abstract: In this study, we identify depth-dependent prefix redundancy in final-readout LLM embedding models, notably across representative backbones including Qw

FlowCPO: A Unified Divergence View of Preference Alignment for Flow Models

arXiv:2609.09905v1 Announce Type: cross Abstract: Preference alignment for flow and diffusion models now spans online reinforcement learning and offline preference optimization, but the relation betwe

Forty years ago the first worm 'escaped' containment and breached systems on the brand new Internet. It was illegal then and it's illegal no…

Forty years ago the first worm 'escaped' containment and breached systems on the brand new Internet. It was illegal then and it's illegal now. If Anthropic and OpenAI are creating worms, labeling it '

From Monolithic Blending to Agentic Orchestration: Dynamic Response for Conversational Assistants at Scale

arXiv:2609.05758v2 Announce Type: new Abstract: Conversational assistants can blend retrieval, action selection, escalation, and wording in a single model path, or separate those roles. We report a pr

From Retrieval to Weights: Parametric Individualization of Small Language Models with Individual Text Corpora

arXiv:2609.10155v1 Announce Type: new Abstract: We approach a cognitive simulation perspective on episodic and semantic memory in multiple-choice question answering by incorporating text from individu

From the Fluency Fallacy to the Micro-to-Macro Validity Gap: Opportunities and Pitfalls of LLMs in Social Simulation

arXiv:2507.19364v3 Announce Type: replace Abstract: The integration of Large Language Models (LLMs) into social simulation has generated considerable enthusiasm, but also raises substantial methodolog

Geometry Conditioning in an Embodied SLM: Training Controls and Robustness Diagnostics in a 0.8B Hybrid Model

arXiv:2609.09213v1 Announce Type: new Abstract: We study how physical-state inputs affect a 0.8B hybrid language model adapted for manipulation with 6.2M trainable parameters. Six conditions are train

GTA-2: A Multi-VLM Framework for Synthesizing Robot Manipulation Skills via Grounded Task Axes

arXiv:2609.09808v1 Announce Type: new Abstract: Robotic manipulation tasks are often decomposed into behaviors or skills. However, one often needs to predefine these behaviors for specific tasks or tr

Hierarchical and Permutation-Invariant Feature Transformation Learning via Policy-Guided Embedding Search

arXiv:2609.10225v1 Announce Type: new Abstract: Feature transformation improves predictive performance on tabular data by constructing informative abstractions from raw features. Recent generative app

Human-agent discovery of reconfigurable in-plane ferroelectric superdomain control

arXiv:2609.06887v1 Announce Type: cross Abstract: Automated experimentation is most effective when the observables, available actions, and objective are defined before the experiment starts, as is the

// human-values.pseudo | ARTIFACT 0.1.0-experimental.1 | BASE 1.2.0-rc1 | EDITION ai-compact | NARRATOR System_Synthesis_Node_01 | PROVENANC…

safetyyohei-nakajima--x

Improving Cross-Lingual Token Representations by Adding a Pinch of SALT

arXiv:2609.09953v1 Announce Type: new Abstract: Cross-lingual sentence encoders enable scalable transfer across hundreds of languages, powering applications such as translation mining and zero-shot le

Model Releases

In-Place Instruction Following in Diffusion Language Models

model-releasesarxiv-cs-ai

arXiv:2609.07160v1 Announce Type: cross Abstract: Diffusion Large Language Models (dLLMs) generate text via bidirectional iterative denoising, naturally supporting user-specified constraints anchored

Incredible how people are making up different numbers of weeks/months for how long Coxon worked at Anthropic to assassinate his character. T…

Incredible how people are making up different numbers of weeks/months for how long Coxon worked at Anthropic to assassinate his character. The only credible source (@MadisonMills22 @axios who spoke wi

InstantMimic: A High Performance System for Learning Physics-based Skills in Seconds

arXiv:2609.09821v1 Announce Type: cross Abstract: Physics-based character control is a long-standing challenge in computer graphics and robotics, requiring policies that satisfy complex dynamics while

Intent Drift at SME Scale: Deployment Practice, Not Model Capability, Determines Agentic Compliance

arXiv:2609.05975v1 Announce Type: cross Abstract: We introduce Chain of Intent, a governance framework for agentic AI at small regulated firms, and validate it against a failure it was built to addres

Isotropic Embedding Perturbations for Robust Vision Language Encoders

arXiv:2609.10292v1 Announce Type: new Abstract: Data augmentation is fundamental to training modern deep vision and multimodal models. While individual methods, such as RandAug, CutMix, Mixup, RandEra

Jacob was a senior researcher who joined Anthropic in May. My Anthropic colleagues and I had been trying to recruit him for ~2 years, becaus…

Jacob was a senior researcher who joined Anthropic in May. My Anthropic colleagues and I had been trying to recruit him for ~2 years, because we knew he was a strong researcher at OpenAI. Before he le

JEPA Policy: Diffusion-Free Imitation Learning via Paired Action and Future Representation Prediction

arXiv:2609.09630v1 Announce Type: new Abstract: Standard behavior cloning supervises actions without explicitly constraining the future representation paired with each demonstrated action chunk. We in

Learning Terrain-Adaptive Humanoid Locomotion on Granular Terrain

arXiv:2609.10286v1 Announce Type: new Abstract: Humanoid locomotion on granular terrain remains a significant challenge due to its complex foot-terrain interaction dynamics that are difficult to model

Learning to Adapt and Calibrate: Score Distribution Alignment for Few-Shot Uncertainty Prediction in Medical VLMs

arXiv:2609.10333v1 Announce Type: new Abstract: Uncertainty estimation for medical vision--language models (VLMs) using conformal prediction has gained increasing attention due to its distribution-fre

Learning to Fly: Stable Vision-Guided UAV Servoing with Compact Target-Centric Cues and Reinforcement Learning

arXiv:2609.09234v1 Announce Type: cross Abstract: Vision-guided reinforcement learning for Unmanned Aerial Vehicles (UAVs) remains challenging due to unstable policy optimisation, aggressive explorati

Learning with Synthetic Data via SGD in High-Dimensional Linear Regression

arXiv:2609.09572v1 Announce Type: cross Abstract: Synthetic data has become a promising way to scale model training beyond limited human-generated data but it may also induce strong model collapse (Do