safety

Sources: the White House's Office of the National Cyber Director and Commerce Department's CAISI are fighting over which agency should lead AI model evaluations (Washington Post)

Washington Post: Sources: the White House's Office of the National Cyber Director and Commerce Department's CAISI are fighting over which agency should lead AI model evaluations — As the White House g

safetytechmeme

Content type

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

Sparse Autoencoders as Plug-and-Play Firewalls for Adversarial Attack Detection in VLMs

arXiv:2605.07447v1 Announce Type: cross Abstract: Vision-language models (VLMs) have advanced rapidly and are increasingly deployed in real-world applications, especially with the rise of agent-based

SparseRL-Sync: Lossless Weight Synchronization with ~100x Less Communication

arXiv:2605.07330v1 Announce Type: cross Abstract: In large-scale reinforcement learning (RL) systems with decoupled Trainer-Rollout execution, the Trainer must regularly synchronize policy weights to

Stabilized neural Hamilton--Jacobi--Bellman solvers: Error analysis and applications in model-based reinforcement learning

arXiv:2605.07116v1 Announce Type: cross Abstract: Physics-informed neural solvers offer a promising route to model-based reinforcement learning in continuous time, where optimal feedback synthesis is

Stationary Reweighting Yields Local Convergence of Soft Fitted Q-Iteration

arXiv:2512.23927v2 Announce Type: replace-cross Abstract: Fitted Q-iteration (FQI) and soft FQI are widely used value-based methods for offline reinforcement learning, but their standard stability gua

STDA-Net: Spectrogram-Based Domain Adaptation for cross-dataset Sleep Stage Classification

arXiv:2605.06736v1 Announce Type: cross Abstract: Accurate sleep stage classification across datasets remains challenging due to variability in EEG channel montages, sampling rates, recording environm

Structured Role-Aware Policy Optimization for Multimodal Reasoning

arXiv:2605.07274v1 Announce Type: new Abstract: Reinforcement learning from verifiable rewards (RLVR), especially with Group Relative Policy Optimization (GRPO), has shown strong potential for improvi

Supervised sparse auto-encoders for interpretable and compositional representations

arXiv:2602.00924v2 Announce Type: replace Abstract: Sparse auto-encoders (SAEs) have re-emerged as a prominent method for mechanistic interpretability, yet they face two significant challenges: the no

Temporal Attention for Adaptive Control of Euler-Lagrange Systems with Unobservable Memory

arXiv:2605.06877v1 Announce Type: new Abstract: Adaptive control of Euler-Lagrange systems is challenging when friction is governed by a finite-horizon internal state that is not directly observable f

Temporal Smoothness Doubly Robust Learning for Debiased Knowledge Tracing

arXiv:2605.05958v2 Announce Type: replace Abstract: Knowledge Tracing (KT) is fundamental to intelligent education systems, yet relies on educational logs that are selectively observed. The non-random

TextLDM: Language Modeling with Continuous Latent Diffusion

arXiv:2605.07748v1 Announce Type: new Abstract: Diffusion Transformers (DiT) trained with flow matching in a VAE latent space have unified visual generation across images and videos. A natural next st

That @NIST CAISI announcement from last week about Google DeepMind, xAI and Microsoft signing new deals for pre-deployment testing is gone f…

That @NIST CAISI announcement from last week about Google DeepMind, xAI and Microsoft signing new deals for pre-deployment testing is gone from their website, link no longer found, and I can't get any

The Cost of Consensus: Malignant Epistemic Herding and Adaptive Gating in Distributed Multi-Agent Search

arXiv:2605.06988v1 Announce Type: cross Abstract: Distributed agents in real-world settings frequently must coordinate under uncertainty with only partial observations. Coordination is necessary to sh

safetyarxiv-cs-ro

The Effect of Mini-Batch Noise on the Implicit Bias of Adam

arXiv:2602.01642v2 Announce Type: replace-cross Abstract: With limited high-quality data and growing compute, multi-epoch training is gaining back its importance across sub-areas of deep learning. Ada

The Endogeneity of Miscalibration: Impossibility and Escape in Scored Reporting

arXiv:2605.07671v1 Announce Type: cross Abstract: Eliciting truthful reports from autonomous agents is a core problem in scalable AI oversight: a principal scores the agent's report using a strictly p

The Limits of AI-Driven Allocation: Optimal Screening under Aleatoric Uncertainty

arXiv:2605.07979v1 Announce Type: new Abstract: The rise of machine learning has shifted targeted resource allocation in policy and humanitarian settings toward algorithmic targeting based on predicte

The Moltbook Files: A Harmless Slopocalypse or Humanity's Last Experiment

arXiv:2605.07462v1 Announce Type: cross Abstract: Moltbook is a Reddit-like platform where OpenClaw agents post, comment, and vote at scale - a so far unprecedented incident that comes with serious sa

The Proxy Presumption: From Semantic Embeddings to Valid Social Measures

arXiv:2605.07409v1 Announce Type: new Abstract: Natural Language Processing is rapidly evolving into a primary instrument for Computational Social Science, with researchers increasingly using embeddin

Theoretical Limits of Language Model Alignment

arXiv:2605.07105v1 Announce Type: cross Abstract: Language model (LM) alignment improves model outputs to reflect human preferences while preserving the capabilities of the base model. The most common

Think-with-Rubrics: From External Evaluator to Internal Reasoning Guidance

arXiv:2605.07461v1 Announce Type: new Abstract: Rubrics have been extensively utilized for evaluating unverifiable, open-ended tasks, with recent research incorporating them into reward systems for re

This is what a useless hype lifecycle looks like.

Gary Marcus critiques the typical hype cycle pattern where emerging technologies experience inflated expectations followed by inevitable disappointment. The post likely illustrates this cycle using a

totally worth $10 trillion a year

This post from AI researcher Gary Marcus likely discusses the enormous economic value or potential return on investment related to artificial intelligence developments, suggesting AI's worth or impact

Toward Better Geometric Representations for Molecule Generative Models

arXiv:2605.07693v1 Announce Type: new Abstract: Geometric representation-conditioned molecule generation provides an effective paradigm that decouples molecule representation modeling from structure g

Towards Differentially Private Reinforcement Learning with General Function Approximation

arXiv:2605.07049v1 Announce Type: cross Abstract: We present the first theoretical guarantees for differentially private online reinforcement learning (RL) with general function approximation, extendi

Towards Fairness under Label Bias in Image Segmentation: Impact, Measurement and Mitigation

arXiv:2605.06891v1 Announce Type: new Abstract: Labeled datasets reflect the biases of their annotation pipelines, which sometimes introduce label bias: group-conditional label errors that cause syste

TRACE: Transport Alignment Conformal Prediction via Diffusion and Flow Matching Models

arXiv:2605.07100v1 Announce Type: cross Abstract: Constructing valid and informative conformal prediction regions for multi-dimensional outputs remains a fundamental challenge. While conformal predict

Training-Free Multimodal Large Language Model Orchestration

arXiv:2508.10016v3 Announce Type: replace Abstract: Building interactive omni-modal assistants often relies on end-to-end multimodal alignment to fuse heterogeneous modalities, which incurs substantia

TRAJGANR: Trajectory-Centric Urban Multimodal Learning via Geospatially Aligned Neural Representations

arXiv:2605.06990v1 Announce Type: new Abstract: Multimodal self-supervised learning (MSSL) has emerged as a key paradigm for pretraining geospatial foundation models. However, existing geospatial MSSL

UFT: Unifying Fine-Tuning of SFT and RLHF/DPO/UNA through a Generalized Implicit Reward Function

arXiv:2410.21438v3 Announce Type: replace Abstract: By pretraining on trillions of tokens, an LLM gains the capability of text generation. However, to enhance its utility and reduce potential harm, SF

UNA: A Unified Supervised Framework for Efficient LLM Alignment Across Feedback Types

arXiv:2408.15339v4 Announce Type: replace-cross Abstract: RL alignment methods, including RLHF and DPO, are primarily based on pairwise preference data. Although scalar or score-based feedback has bee

UniD-Shift: Towards Unified Semantic Segmentation via Interpretable Share-Private Multimodal Decomposition

arXiv:2605.07356v1 Announce Type: new Abstract: Semantic segmentation of large-scale 3D point clouds is crucial for applications such as autonomous driving and urban digital twins. However, the sparse

VDEGaussian: Video Diffusion Enhanced 4D Gaussian Splatting for Dynamic Urban Scenes Modeling

arXiv:2508.02129v2 Announce Type: replace Abstract: Dynamic urban scene modeling is a rapidly evolving area with broad applications. While current approaches leveraging neural radiance fields or Gauss

VESPO: Variational Sequence-Level Soft Policy Optimization for Stable Off-Policy LLM Training

arXiv:2602.10693v3 Announce Type: replace-cross Abstract: Off-policy updates are inevitable in reinforcement learning (RL) for large language models (LLMs) due to rollout staleness from asynchronous t

VideoRouter: Query-Adaptive Dual Routing for Efficient Long-Video Understanding

arXiv:2605.05848v2 Announce Type: replace-cross Abstract: Video large multimodal models increasingly face a scalability bottleneck: long videos produce excessively long visual-token sequences, which s

VISD: Enhancing Video Reasoning via Structured Self-Distillation

arXiv:2605.06094v2 Announce Type: replace-cross Abstract: Training VideoLLMs for complex reasoning remains challenging due to sparse sequence level rewards and the lack of fine grained credit assignme

what i have been saying for 6 years. maybe now you will believe it?

what i have been saying for 6 years. maybe now you will believe it? 📁 Fei-Fei Li, former Google Chief Scientist, says the industry is dangerously fixated on language models. Most of the real economy i

When Descent Is Too Stable: Event-Triggered Hamiltonian Learning to Optimize

arXiv:2605.06868v1 Announce Type: new Abstract: Fixed-budget nonconvex optimization can fail not because local descent is unstable, but because it is too stable: after reaching a nearby stationary poi

Which has better odds? Generative AI earning $1.6 trillion/year or the roulette wheel landing on zero?

Which has better odds? Generative AI earning $1.6 trillion/year or the roulette wheel landing on zero? One estimate of how much annual revenue AI needs to “make sense”: 1.6 trillion. That’s four times

Who Prices Cognitive Labor in the Age of Agents? Compute-Anchored Wages

arXiv:2605.05558v2 Announce Type: replace Abstract: A natural intuition about the economics of AI agents is that, because agents can be replicated at very low marginal cost, agent labor may be supplie

Why Does Agentic Safety Fail to Generalize Across Tasks?

arXiv:2605.06992v1 Announce Type: new Abstract: AI agents are increasingly deployed in multi-task settings, where the task to perform is specified at test time, and the agent must generalize to unseen

Zero-Shot Imagined Speech Decoding via Imagined-to-Listened MEG Mapping

arXiv:2605.08075v1 Announce Type: new Abstract: Decoding imagined speech from non-invasive brain recordings is challenging because imagined datasets are scarce and difficult to align temporally across

32,000 views for my tweet; 2.2 million for the nonsense it critiques. so typical. lies and distortions beating debunking by a factor of roug…

32,000 views for my tweet; 2.2 million for the nonsense it critiques. so typical. lies and distortions beating debunking by a factor of roughly 100. Wanna get a million views? Make stuff up. Take a ti

a tweet to keep in mind when Sam testifies this coming week

a tweet to keep in mind when Sam testifies this coming week How can anybody take seriously your claim that “Working towards prosperity for everyone, empowering all people, and advancing science and te

Big AI Lobbyists: if you regulate us at all, we lose to China because they will never regulate ... Actual China: 'safety first, innovation second ... Development must be controllable and orderly.'

This post highlights a contradiction in AI industry arguments, contrasting Western tech company claims that regulation will disadvantage them competitively against China with evidence of China's own s

safetyr-chatgpt

FAR more scary than the misunderstood METR time horizon graph everyone here seems to be (wrongly) panicking about.

FAR more scary than the misunderstood METR time horizon graph everyone here seems to be (wrongly) panicking about. The #AI circular funding bubble is already twice the size of the outstanding debt in

Happy to put money against superintelligence in 2029.

Happy to put money against superintelligence in 2029. Why Superintelligence is Alien Intelligence (A Brief Outline of the Future) We are no longer in the realm of normal technological progress. If the

hey @elonmusk, if you really care about making X a source of truth as you used to claim, take note:

hey @elonmusk, if you really care about making X a source of truth as you used to claim, take note: @GaryMarcus lies are optimized for engagement, truth is optimized for accuracy the algorithm doesnt

Many are valiantly fighting against the age of bullshit. Bullshit is winning, I'm sorry to report. 😠

Many are valiantly fighting against the age of bullshit. Bullshit is winning, I'm sorry to report. 😠 32,000 views for my tweet; 2.2 million for the nonsense it critiques. so typical. lies and distorti