safety

The Role of Feedback Alignment in Self-Distillation

arXiv:2606.11173v1 Announce Type: new Abstract: Conditioning a language model on additional context, such as feedback on a previous attempt, typically improves its response. Self-distillation trains t

The Whale That Outswam Evolution: Swarm Intelligence Maximises Memory in Connectome Reservoirs

Content type

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

DGX agent

arXiv:2606.09902v1 Announce Type: cross Abstract: Reservoir computing exploits the fixed dynamics of a recurrent network for temporal processing, requiring only a trained linear readout. Biological ne

safetyarxiv-cs-ai

10 Jun 2026

there is an epidemic of this scam, with stolen picture and random user names and handles. note my reply lol and do not get taken.

Gary Marcus warns about a widespread scam epidemic involving fake profiles that use stolen pictures and randomly generated usernames/handles to deceive people. He shares an example of his response to

This is genuinely big news, major unintended consequences.

This is genuinely big news, major unintended consequences. 🚨Breaking news that could be huge, and enormously bad for GenAI, if other countries make similar decisions. https://the-decoder.com/landmark-

this is important. and scary.

this is important. and scary. CAISI has reportedly been directed to stop publishing public model assessments as the new AI EO gets implemented. Natsec engagement on AI is essential. But pulling CAISI'

Today I'm publishing a new essay, Policy on the AI Exponential. AI is progressing extremely fast—much faster than the policy process was bui…

Today I'm publishing a new essay, Policy on the AI Exponential. AI is progressing extremely fast—much faster than the policy process was built to handle. The essay lays out where I think the technolog

safetyboris-cherny--x

Toward Calibrated, Fair, and accurate Deepfake Detection

arXiv:2606.09881v1 Announce Type: cross Abstract: Deepfake detectors show large performance gaps across demographic groups. Existing fairness approaches require demographic labels, retraining, or sacr

TRACE: A Unified Rollout Budget Allocation Framework for Efficient Agentic Reinforcement Learning

arXiv:2606.11119v1 Announce Type: cross Abstract: Reinforcement learning with verifiable rewards (RLVR) is a promising approach for enhancing reasoning and agentic behavior in large language models. H

Trading Utility for Dynamic Fairness in Multiple Resource Division with Sequential Demand

arXiv:2606.10472v1 Announce Type: cross Abstract: Dynamic multi-resource allocation is a central problem in shared computing environments, where users' demands arrive sequentially and resources must b

safetyarxiv-cs-lg

Two to Tango: Coupled Task-Reference Selection for Safe LLM Fine-tuning

arXiv:2606.09866v1 Announce Type: cross Abstract: Fine-tuning safety aligned large language models (LLMs) on downstream data improves adaptation but may erode learned safety behavior. Existing methods

Uncovering Vulnerability of Vision-Language-Action Models under Joint-Level Physical Faults

arXiv:2606.10501v1 Announce Type: new Abstract: Deploying Vision-Language-Action (VLA) models in real robotic systems requires robustness not only to semantic and perceptual variations, but also to em

safetyarxiv-cs-ro

UniPET: a universal network for high-quality PET image denoising across varied dose reduction factors

arXiv:2606.11131v1 Announce Type: new Abstract: Most existing deep learning-based PET image denoising methods assume a fixed and known dose reduction factor (DRF) for low-dose PET images. However, the

Using Probabilistic Programs to Train Inductive Reasoning in Large Language Models

arXiv:2606.09856v1 Announce Type: cross Abstract: Post-training Large Language Models (LLMs) for reasoning typically focuses on deductive tasks such as mathematics and coding where correctness is veri

Visual-TCAV: Concept-based Attribution and Saliency Maps for Post-hoc Explainability in Image Classification

arXiv:2411.05698v3 Announce Type: replace-cross Abstract: Convolutional Neural Networks (CNNs) have shown remarkable performance in image classification. However, interpreting their predictions is cha

Warren to SEC, lightly paraphrased: “Do your f’ing job, and don’t let retail investors get screwed”

Senator Elizabeth Warren criticized the SEC for insufficient enforcement and investor protection, urging the agency to strengthen oversight and prevent harm to retail investors. The post, shared by AI

What if, Germany locked itself out of the LLM race and • Had students who actually learned things in high school, instead of turning in prom…

What if, Germany locked itself out of the LLM race and • Had students who actually learned things in high school, instead of turning in prompt outputs they barely read • Emerged from the sea of slop •

What Should a Skill Remember? Quality--Cost Trade-offs in Cost-Aware Skill Rewriting for Language Model Agents

arXiv:2606.09421v2 Announce Type: replace Abstract: Large language model agents increasingly rely on skills: reusable procedural documents encoding workflows, tool use, implementation patterns, valida

safetyarxiv-cs-cl

When Distance Distracts: Representation Distance Bias in BT-Loss for Reward Models

arXiv:2512.06343v3 Announce Type: replace-cross Abstract: Reward models are central to Large Language Model (LLM) alignment within the framework of RLHF. The standard objective used in reward modeling

When the Chain of Thought Knows Better: Failure Modes in Multi-Turn Reasoning Models

arXiv:2606.10740v1 Announce Type: new Abstract: Failures in multi-turn reasoning models are largely invisible to terminal-score evaluation. A model can lock onto an unsafe stance early in a long dialo

When to Align, When to Predict: A Phase Diagram for Multimodal Learning

arXiv:2606.11190v1 Announce Type: new Abstract: Cross-modal alignment (CA) and cross-modal prediction (CP) are the dominant paradigms for multimodal representation learning, yet there is no systematic

safetyarxiv-cs-lg

When you hear AI 'safety' you should hear 'censorship' and 'control' instead. All of us surveilled and spied by safeguards of loving grace. …

When you hear AI 'safety' you should hear 'censorship' and 'control' instead. All of us surveilled and spied by safeguards of loving grace. Today it's intelligent Terms of Service control. You can't d

safetyyann-lecun--x

Why are AI research restrictions treated differently from every other safeguard? @theemozilla and @karan4d, co-founders of Nous Research: 'O…

Why are AI research restrictions treated differently from every other safeguard? @theemozilla and @karan4d, co-founders of Nous Research: 'On the bio stuff... just saying no to the user and being hone

safetynous-research--x

Wow. This is worth watching, might have huge impact. @SenWarren makes some valid points, and the SEC owes her public answers

Wow. This is worth watching, might have huge impact. @SenWarren makes some valid points, and the SEC owes her public answers Sen. Warren calls on SEC to delay SpaceX IPO @CNBC https://www.cnbc.com/202

YUBI: Yielding Universal Bidigital Interface for Bimanual Dexterous Manipulation at Scale

arXiv:2606.10244v1 Announce Type: cross Abstract: We introduce Yielding Universal Bidigital Interface (YUBI), a finger-aligned gripper designed to enable intuitive, ergonomic, and scalable data collec

6G Empowering Future Robotics: A Vision for Next-Generation Autonomous Systems

arXiv:2602.12246v2 Announce Type: replace-cross Abstract: The convergence of robotics and next-generation communication is a critical driver of technological advancement. As the world transitions from

safetyarxiv-cs-ro

A Finetuned SpeechLLM for Joint Multi-Granular L2 Assessment and Natural-Language Rationales

arXiv:2606.09470v1 Announce Type: cross Abstract: Automated L2 speech assessment can assign proficiency labels, but often lacks interpretability. We propose a rubric-guided SpeechLLM for multi-aspect,

A Geometric Unification of Concept Learning with Concept Cones

arXiv:2512.07355v2 Announce Type: replace Abstract: Two traditions of interpretability have evolved side by side but seldom spoken to each other: Concept Bottleneck Models (CBMs), which prescribe what

A Joint Finite-Sample Certificate for Adaptive Selective Conformal Risk Control

arXiv:2606.08517v1 Announce Type: new Abstract: Selective predictors answer on confident inputs and abstain elsewhere; deploying one safely needs a single finite-sample certificate that simultaneously

safetyarxiv-cs-lg

A Mixed Diet Makes DINO An Omnivorous Vision Encoder

arXiv:2602.24181v2 Announce Type: replace-cross Abstract: Pre-trained vision encoders like DINOv2 have demonstrated exceptional performance on unimodal tasks. However, we observe that their features a

A practical probabilistic framework for deformable image registration uncertainty in radiotherapy dose propagation

arXiv:2606.09253v1 Announce Type: new Abstract: Deformable image registration (DIR) is widely used in radiotherapy for dose propagation and accumulation, but uncertainty in the underlying deformation

A Unifying Lens on Reward Uncertainty in RLHF

arXiv:2606.09073v1 Announce Type: cross Abstract: Reinforcement learning from human feedback (RLHF) is bottlenecked by reward hacking, where the policy exploits errors in a proxy reward model (RM) and

A VideoMAE-v2 Approach to Zero-Shot Traffic Accident Anticipation

arXiv:2606.09542v1 Announce Type: new Abstract: Traffic accident anticipation -- predicting the likelihood of an imminent collision at every frame of a dashcam video -- is safety-critical yet difficul

Ablation-Reversible Heads Don't Transfer: A Stress Test for Mechanistic Role Claims in Transformers

arXiv:2606.08292v1 Announce Type: new Abstract: In mechanistic interpretability, attention heads are commonly elevated to role claims (e.g., 'this head represents addition') when they are necessary fo

absolutely correct

absolutely correct @GaryMarcus I've recently learned not to believe anything that anyone says anymore when it comes to AI companies. They're all competing to win against each other while exploiting, s

ActProbe: Action-Space Probe for Early Failure Detection of Generative Robot Policies

arXiv:2606.08508v1 Announce Type: cross Abstract: Generative robot policies fail unpredictably at deployment: they hesitate at critical moments, drift off-task, or commit to unrecoverable actions. Exi

Adaptive Loss Balancing for Noise-Robust GRPO in Generative Recommendation

arXiv:2606.08480v1 Announce Type: cross Abstract: Reinforcement learning (RL) presents a promising avenue for enhancing generative recommendation beyond supervised imitation, leveraging reward signals

AeroSpectra Sentinel: An Auditable LLM Prompt-Chaining Decision-Support Workflow for Acute Asthma Risk Assessment from Respiratory Sounds and Clinical Signals

arXiv:2606.08247v1 Announce Type: cross Abstract: Acute asthma risk assessment requires rapid interpretation of respiratory sounds, oxygenation, airflow limitation, speech ability, work of breathing,

Agent Economics: An Entropy-Controlled Pluralistic Alignment Framework for Preventing Artificial Hivemind in Autonomous Agents

arXiv:2606.09039v1 Announce Type: new Abstract: This study proposes the Behavioral Protocol Framework (BPF), an entropy-controlled pluralistic alignment framework designed to address two critical chal

AgriGov: A Structured Multilingual Dataset Curation for Indian Government Schemes for Farmers

arXiv:2606.08272v1 Announce Type: cross Abstract: AgriGov is a curated, trilingual (English-Hindi-Marathi) dataset designed to address the scarcity of domain-grounded multilingual resources for agricu

AHA-WAM:Asynchronous Horizon-Adaptive World-Action Modeling with Observation-Guided Context Routing

arXiv:2606.09811v1 Announce Type: cross Abstract: World-action models have emerged as a promising paradigm for robot manipulation, jointly modeling visual scene dynamics and actions to inject physical

AI Assurance in UK Defence: Challenges in Operationalising JSP 936

arXiv:2606.09414v1 Announce Type: cross Abstract: This report examines practical challenges in operationalising JSP 936 Part 1 for AI assurance in UK Defence. Using a structured interpretive review of

AI Code Sandboxes: A Comparative Security Study. Part 1 of 2 -- Engine-Level Properties (Attack Surface, Leakage, Stackability, CVE History, Patch Cadence, Fuzzing)

arXiv:2606.08433v1 Announce Type: cross Abstract: This paper reads six engine-level measurements together -- 1.1 host attack surface, 1.2 information leakage, 1.3 defense-in-depth stackability, 1.4 pu

AI-Integrated Learning Management System for Middle School: A Longitudinal Study of Learning Outcomes Through High School and Beyond

arXiv:2606.07544v1 Announce Type: cross Abstract: Middle school is a key window for building core academic skills and the learning routines students carry into later grades, yet many students still fa

alienating your customers before you IPO is maybe not the best idea, @AnthropicAI

alienating your customers before you IPO is maybe not the best idea, @AnthropicAI Brilliant idea! Next up: Apple randomly reboots your Mac if you're building competing tech, Gmail silently edits your

Aligned but Not Partner-Specific: Distinguishing How Multimodal LLM Agents Succeed in Reference Games Without Human-Like Conventions

arXiv:2606.08081v1 Announce Type: cross Abstract: Repeated reference games test whether interlocutors replace their initially long descriptions with shorter, partner-specific conventions grounded in s

AMix-1: A Pathway to Test-Time Scalable Protein Foundation Model

arXiv:2507.08920v4 Announce Type: replace-cross Abstract: We introduce AMix-1, a powerful protein foundation model built on Bayesian Flow Networks and empowered by a systematic training methodology, e

An Agency-Transferring Model-Free Policy Enhancement Technique

arXiv:2606.09825v1 Announce Type: cross Abstract: Training reinforcement learning (RL) policies from scratch is costly: it requires careful reward and environment design, extensive tuning, and substan

Anchor-Conditioned Compositional Control for Landscape Image Generation

arXiv:2606.07638v1 Announce Type: cross Abstract: Image generative models, though widely used as creative tools, offer limited support for the kind of compositional control that photographers and visu