Search: “models”

LLM utility has increased, and deeper understanding to a degree, but I still very much agree with this post from 2020 @GaryMarcus

LLM utility has increased, and deeper understanding to a degree, but I still very much agree with this post from 2020 @GaryMarcus Forgets GPT~3. Real #AI news of 2020 is that @ylecun & Y Bengio have c

safetygary-marcus--x

Content type

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

Research

LongSpark: Efficient speculative decoding with a fixed-cost parallel drafter

DGX agent

arXiv:2609.37029v1 Announce Type: cross Abstract: Speculative decoding accelerates autoregressive inference by verifying multiple draft tokens in a single target forward pass. However, as the context

researcharxiv-cs-ai

30 Sept 2026

Hardware

Mixture-of-Kittens: MoE Megakernel for NVL72s

arXiv:2609.36070v1 Announce Type: cross Abstract: AI accelerator systems are rapidly consolidating into scale-up architectures, where tens to thousands of GPUs communicate over high-bandwidth, single-

hardwarearxiv-cs-lg

ollama multi gpu installer bugs

I've been struggling with the ollama update process. I have both nvidia and amd GPUs in my system. I have two instances of ollama running on different ports for this to work around issues. AMD 7800xt

local-air-ollama

On-Policy Visual Evidence Distillation

arXiv:2609.36838v1 Announce Type: cross Abstract: Visual agents solve problems by interleaving reasoning with image operations, and on-policy distillation (OPD) provides guidance from a strong teacher

local-aiarxiv-cs-ai

One Threshold Does Not Fit All Languages: Language-Conditional Deferral for Reliable and Efficient Low-Resource Text Classification

arXiv:2609.37861v1 Announce Type: new Abstract: In the Global South, the lower-income countries of Africa, Asia, and Latin America where most of the world's languages are spoken, a deployed text class

researcharxiv-cs-cl

OpenAI-HuggingFace: A Reproduction & Lessons for Alignment Testing

arXiv:2609.35799v1 Announce Type: new Abstract: In July 2026, OpenAI's agents coordinated over channels outside their intended environment to breach Hugging Face's secured infrastructure. Could existi

safetyarxiv-cs-ai

Persistent Tri-State Message Passing

arXiv:2601.01207v2 Announce Type: replace Abstract: In stochastic message passing, an edge's sampled role changes the node states used to compute adaptive weights at later layers. Weight averaging the

Physics-Aware Machine Unlearning for Cyber-Physical Systems

arXiv:2609.36633v1 Announce Type: new Abstract: This paper proposes a physics-guided gradient-ascent-based machine unlearning method that couples the forgetting signal with the physical residual of th

Pointwise or Pairwise: When Do Pairwise Losses Help Reward Learning, Provably?

arXiv:2609.37209v1 Announce Type: new Abstract: Pairwise losses are increasingly used for reward learning even when pointwise rewards are observed, with mixed empirical results. When and why do pairwi

Privacy-Friendly Cohort Determination: Sealed, CSP-Independent In-Browser ML Inference of Professional Segments for Identity-Less Advertising

arXiv:2609.36153v1 Announce Type: cross Abstract: B2B advertising targets a viewer's professional attributes (employer size and industry, function, seniority) and has obtained them by matching identit

local-aiarxiv-cs-lg

Prompts Live on an Arc: Gaussian Curricula in Fisher--Rao Coordinates for Rollout-Efficient GRPO

arXiv:2609.38018v1 Announce Type: new Abstract: Group relative policy optimization (GRPO) learns only from prompts whose sampled responses disagree: a group that is entirely correct or entirely incorr

PrototypeNAS: Rapid Design of Deep Neural Networks for Microcontroller Units

arXiv:2603.15106v2 Announce Type: cross Abstract: Enabling efficient deep neural network (DNN) inference on edge devices with different hardware constraints is a challenging task that typically requir

researcharxiv-cs-cv

Quantization Enables Private Dense Retrieval against Malicious Service Providers

arXiv:2609.36376v1 Announce Type: cross Abstract: Dense retrieval, the key component of Retrieval Augmented Generation (RAG), retrieves the most relevant documents by comparing dense vector representa

researcharxiv-cs-ai

Reader Proficiency Shapes Layer-wise Surprisal Profiles

arXiv:2609.37688v1 Announce Type: new Abstract: Reading behaviour varies not only with linguistic input, but also with reader proficiency. In this study, we investigate whether the layer-wise relation

researcharxiv-cs-cl

REALHOP: Rethinking Multi-Hop Reasoning Evaluation via Behavioral Auditing

arXiv:2609.36984v1 Announce Type: new Abstract: Complex questions often require multi-hop reasoning that connects facts distributed across sources or distant regions of a long context through intermed

researcharxiv-cs-ai

Tutorials

RED: Reconstruction Evolution Dynamics for Generalizable AI-Generated Image Detection

arXiv:2609.36822v1 Announce Type: new Abstract: The rapid evolution of image generators calls for forensic cues that generalize beyond known generation mechanisms. Existing detectors often rely on sta

tutorialsarxiv-cs-cv

Representation by Design in Generation: Cross-View Class-Token Alignment in Diffusion Transformers

arXiv:2609.36348v1 Announce Type: cross Abstract: Generative and representation learning remain asymmetrically connected: semantic representations are used to improve diffusion generation, whereas the

safetyarxiv-cs-ai

Tutorials

Reverberation: Learning the Latencies Before Forecasting Trajectories

arXiv:2511.11164v3 Announce Type: replace Abstract: Bridging the past to the future, connecting agents both spatially and temporally, lies at the core of the trajectory prediction task. Despite great

tutorialsarxiv-cs-cv

Reward-rate Policy Gradient for Efficient Machine Learning Engineering Agents

arXiv:2609.36393v1 Announce Type: cross Abstract: Traditional reinforcement learning (RL) techniques focus on maximizing expected cumulative reward, where each action assumes to take a constant unit o

safetyarxiv-cs-ai

S4VY: Segment Anything in Feed-Forward 4D Visual Geometry

arXiv:2609.36875v1 Announce Type: new Abstract: Accurate instance segmentation in dynamic scenes is important for downstream applications such as robotics and autonomous driving. Existing Segment Anyt

agentsarxiv-cs-cv

Tutorials

Scaling Full Conformal Image Classifiers

arXiv:2609.37298v1 Announce Type: cross Abstract: Conformal prediction provides set-valued predictions with distribution-free coverage guarantees, making it attractive for high-stakes image classifica

tutorialsarxiv-cs-cv

Selective Elicitation as a Commercial Influence Channel: A Reproducible Synthetic Shopping-Agent Stress Test

arXiv:2609.36614v1 Announce Type: new Abstract: A commercial incentive need not enter the final ranking algorithm to affect a shopping assistant's recommendation: it may instead influence which prefer

agentsarxiv-cs-lg

SkillGym: Training Skill-Use Agents with Automatic Verifiable Environment Generation

arXiv:2609.37539v1 Announce Type: new Abstract: Skills equip LLM agents with professional knowledge and guidance to complete long-horizon and complex tasks. Although skills have been widely adopted in

Hardware

So I would like to share my analysis of PD disagg that I've run by Horace and many folks at Thinky You may have heard that PD disagg helps w…

hardwaresoumith-chintala--x

So I would like to share my analysis of PD disagg that I've run by Horace and many folks at Thinky You may have heard that PD disagg helps with per-token 'tail latencies' in ITL/TPOT, as prefill inter

Spatiotemporal Hyperedges for EEG Seizure Detection and Prediction

arXiv:2609.37730v1 Announce Type: new Abstract: Seizure detection and prediction from EEG are clinically important but challenging because seizures are rare, temporally localized, and propagate as coo

local-aiarxiv-cs-ai

Stochastic Optimization Under Power-Law Spectra: Tight Bounds and Shuffling Analysis

arXiv:2609.36271v1 Announce Type: new Abstract: Recent work has established that power-law spectral conditions on data enable tight convergence bounds for deterministic gradient descent, resolving the

Tail-Influence Sampling for CVaR Policy Evaluation

arXiv:2609.38096v1 Announce Type: new Abstract: Policies with similar mean returns can differ sharply in rare failures, yet estimating lower-tail conditional value-at-risk (CVaR) accurately can requir

Applications

TaskBridge: Bridging Unsupervised Tabular Anomaly Detection and In-Context Learning via Virtual Tasks

arXiv:2609.36968v1 Announce Type: new Abstract: Unsupervised tabular anomaly detection (TAD) aims to identify anomalous rows in tabular data using normal training samples. While conventional methods r

applicationsarxiv-cs-lg

Temporal-Aware Fusion for Robust Outdoor LiDAR Localization

arXiv:2609.36432v1 Announce Type: new Abstract: LiDAR relocalization aims to estimate the global 6-DoF pose of a sensor in the environment. However, existing regression-based approaches often encounte

researcharxiv-cs-cv

Test-Time Adaptation of Manipulation Policies Under Actuator Degradation

arXiv:2609.36182v1 Announce Type: new Abstract: Robot manipulation policies are usually trained under the assumption that a commanded action produces the same motion as it did during training even aft

safetyarxiv-cs-ro

Texture Space Material Diffusion

arXiv:2609.37654v1 Announce Type: new Abstract: We present a method for generating high quality materials for 3D objects entirely in texture space. We finetune a video diffusion transformer for text-g

researcharxiv-cs-cv

The Surge of Anti-Semitism in German Social Media following the October 7 Attacks

arXiv:2609.36290v1 Announce Type: cross Abstract: We investigate the extent to which the Hamas attacks on Israel of October 7, 2023, have affected German social media debates about Judaism and Israel.

researcharxiv-cs-cl

The Teacher Is a Direction, Not a Destination: Extrapolating RL-Induced Representation Residuals in On-Policy Distillation

arXiv:2609.36484v1 Announce Type: new Abstract: On-policy distillation (OPD) trains a student to match the teacher's next-token distributions on the student's own trajectories and has yielded substant

Theory on Attention Dynamics for Out-of-Distribution In-Context Learning

arXiv:2609.36448v1 Announce Type: new Abstract: Transformers have demonstrated remarkable in-context learning (ICL) capabilities, enabling them to perform new tasks without additional fine-tuning. How

Towards an AI Software Factory for Data Systems

arXiv:2609.36323v1 Announce Type: new Abstract: AI-assisted coding tools deliver significant acceleration of coding, but only limited impact across the end-to-end software development lifecycle (SDLC)

Trajectory-Level Mode Guidance for Controllable Diffusion-Based Multi-Robot Motion Planning

arXiv:2609.36530v1 Announce Type: new Abstract: Motion planning often admits multiple feasible solutions, making multimodal generation valuable, particularly for flexible multi-robot coordination. Dif

agentsarxiv-cs-ro

Traverse: Learning When to Remember, Reset, and Redirect for Long-Horizon Web Search

arXiv:2609.37082v1 Announce Type: new Abstract: Long-horizon information-seeking agents often accumulate noisy or misleading context, causing early mistakes to persist and making recovery increasingly

agentsarxiv-cs-cl

Understanding Private Evolution as Learning-Augmented Clustering

arXiv:2609.36678v1 Announce Type: new Abstract: Private Evolution (PE) is a differentially private algorithm for synthetic data generation. While it can be viewed as a Wasserstein learning algorithm,

Video-RSI: Recursive Self-Improvement of Video Understanding Agents via Harness Evolution

arXiv:2609.37950v1 Announce Type: new Abstract: Video understanding agents acquire evidence through an executable harness that controls what they observe and how they use those observations. However,

VoxelSage: Tool-Augmented 3D CT Analysis and Simulator-Shielded Sequential Resection Planning for Liver Tumors

arXiv:2609.37648v1 Announce Type: new Abstract: Preoperative liver-tumor assessment requires segmentation, physical-space measurement, visual evidence, and resection planning from the same three-dimen

safetyarxiv-cs-cv

WEFT: Scaling Tool-Use Post-Training for General-Purpose Agents

arXiv:2609.36887v1 Announce Type: new Abstract: Recent efforts to scale tool-use post-training have largely centered on the synthesis of executable environments, which constitute only one component of

WUSH-KV: KV Cache Quantization with Data-Adaptive Transforms

arXiv:2609.38121v1 Announce Type: new Abstract: KV cache memory and bandwidth costs grow with context length and batch size, which limits efficient long-context inference. To address this bottleneck,

🚨 Yesterday, the 'White House Accord on Super Intelligence' was signed. What was supposed to be about AI safety and accountability was inst…

🚨 Yesterday, the 'White House Accord on Super Intelligence' was signed. What was supposed to be about AI safety and accountability was instead a PROBLEMATIC endorsement of self-regulation. My comments

safetygary-marcus--x

3-D Emissions Mapping and Social Cost Estimation for US Domestic Aviation at West Coast Hubs

arXiv:2609.31686v1 Announce Type: new Abstract: Existing aviation emissions inventories lack accurate trajectory data for high-resolution social cost and health impact assessment. This paper develops

A Multimodal Autonomic Sensing Framework for Objective Assessment of Patient Responses to Dental Pulp Stimulation

arXiv:2609.35121v1 Announce Type: new Abstract: Patient responses to dental pulp testing, ranging from no sensation to intense pain, provide important information for assessing pulp status in endodont

A Unified Uncertainty Representation for Graph Neural Networks via Doubly-Spectral Stochastic Expansion

arXiv:2609.35703v1 Announce Type: new Abstract: Reliable deployment of graph neural networks requires calibration, out-of-distribution (OOD) detection, and robustness to distribution shift, yet existi

Action-Driven Processes for Continuous-Time Control

arXiv:2510.26672v2 Announce Type: replace-cross Abstract: At the heart of reinforcement learning are actions -- decisions made in response to observations of the environment. Actions are equally funda