Search: “safety”

Semantic Early-Stopping for Iterative LLM Agent Loops

arXiv:2606.27009v1 Announce Type: new Abstract: Multi-agent large language model (LLM) loops, for example a Writer that drafts and a Critic that revises, are almost always terminated by a fixed iterat

Sketched Linear Contrastive Learning: Approximation, Optimization, and Statistical Scaling

Content type

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

DGX agent

arXiv:2606.26617v1 Announce Type: new Abstract: Scaling laws describe how learning performance varies with model size, data size, and compute. While recent theoretical work has established scaling law

safetyarxiv-cs-lg

26 Jun 2026

Soft Token Alignment for Cross-Lingual Reasoning

arXiv:2606.26466v1 Announce Type: new Abstract: Multilingual large language models often produce inconsistent reasoning and answers for semantically equivalent prompts in different languages. Prior wo

Somatic in the East, Psychological in the West?: Investigating Clinically-Grounded Cross-Cultural Depression Symptom Expression in LLMs

arXiv:2508.03247v2 Announce Type: replace Abstract: Prior clinical psychology research shows that Western individuals with depression tend to report psychological symptoms, while Eastern individuals r

source: “How much compute does the world really need? Scale cannot solve AI’s fundamental problem with accuracy” https://www.ft.com/content/…

This article argues that increasing computational scale alone cannot address fundamental accuracy and reliability problems in AI systems, suggesting that AI development requires solutions beyond raw c

SpatialFlow-GRPO: Where Spatial Credit Drives Image Editing

arXiv:2606.26872v1 Announce Type: new Abstract: Recent online reinforcement learning has substantially improved image editing quality. However, existing Flow-GRPO-style methods usually rely on a singl

SPCX - SPACEX BOND SELLOFF DEEPENS SpaceX's 25 billion bond sale is suffering unusually steep losses, with paper losses exceeding $300 mil…

SPCX - SPACEX BOND SELLOFF DEEPENS SpaceX's 25 billion bond sale is suffering unusually steep losses, with paper losses exceeding $300 million. Traders say fast-money investors may be exiting, while c

$SPCX falls below 150. If you bought it once it the market you lost money. If you bought at 225 you lost a third in less than two weeks. Non…

Gary Marcus comments on significant losses in $SPCX stock, noting that investors who purchased at 225 experienced approximately one-third losses within two weeks as the stock fell below 150. The post

State Representation Matters in Deep Reinforcement Learning: Application to Energy Trading

arXiv:2606.27032v1 Announce Type: cross Abstract: Energy trading decisions depend not only on current market prices, but also on expected future market conditions, and operational constraints. This ma

Statistical and Structural Approaches to Algorithmic Fairness

arXiv:2606.26200v1 Announce Type: cross Abstract: Modern machine learning systems have outgrown their origins as isolated predictive constructs, evolving into complex socio-technical architectures tha

Staying VIGILant: Mitigating Visual Laziness via Counterfactual Visual Alignment in MLLMs

arXiv:2606.26387v1 Announce Type: cross Abstract: Multimodal large language models (MLLMs) extend large language models (LLMs) with visual perception, enabling joint reasoning over images and text. De

STORM: Slot-based Task-aware Object-centric Representation for robotic Manipulation

arXiv:2601.20381v3 Announce Type: replace Abstract: Visual foundation models provide strong perceptual features for robotics, but their dense representations lack explicit object-level structure, limi

SymQNet: Amortized Acquisition for Low-Latency Adaptive Hamiltonian Learning

arXiv:2606.12808v3 Announce Type: replace-cross Abstract: Adaptive Hamiltonian learning is central to calibrating and characterizing quantum devices. In an adaptive controller, choosing the next exper

Tactile-WAM: Touch-Aware World Action Model with Tactile Asymmetric Attention

arXiv:2606.26663v1 Announce Type: new Abstract: World Action Models (WAMs) generate actions together with predicted futures, offering a powerful interface for robot decision making. In contact-rich ma

The Verification Horizon: No Silver Bullet for Coding Agent Rewards

arXiv:2606.26300v1 Announce Type: new Abstract: A classical intuition holds that verifying a solution is easier than producing one. For today's coding agents, this intuition is being inverted: as foun

Things came undone this week: Starmer resigned, SpaceX tanked, the Iran deal stumbled, and a housing bill that passed the Senate is dying on…

Things came undone this week: Starmer resigned, SpaceX tanked, the Iran deal stumbled, and a housing bill that passed the Senate is dying on Trump’s desk. Today’s conversation with @GaryMarcus is a mu

This is what’s causing Anthropic to aggressively beg for govt protection (see below). Customers are finding cheaper alternatives. Keeping em…

This is what’s causing Anthropic to aggressively beg for govt protection (see below). Customers are finding cheaper alternatives. Keeping employees requires continuing ultra-rich secondaries ($$) that

Training-Free Generation of Protein Sequences from Small Family Alignments via Stochastic Attention

arXiv:2603.14717v2 Announce Type: replace Abstract: Generating novel protein sequences that respect a family's statistical constraints typically requires training deep generative models on thousands t

Understanding Domain-Aware Distribution Alignment in Budgeted Entity Matching

arXiv:2606.27342v1 Announce Type: cross Abstract: Entity Matching (EM) is a core operation in the data integration pipeline, where records from different sources are compared to determine whether they

VibeAct: Vibration to Actions for Contact-Rich Reactive Robot Dexterity

arXiv:2606.27344v1 Announce Type: new Abstract: Dexterous manipulation depends on contact events that are fast, local, and often visually occluded. Piezoelectric microphones offer a compact and high-b

Visual-OPSD: Cross-Modal On-Policy Self-Distillation for Efficient Unified Multimodal Reasoning

arXiv:2606.18974v2 Announce Type: replace Abstract: Unified multimodal models (UMMs) interleave generated ''visual thoughts'' (VTs) with text reasoning to improve spatial tasks. This incurs roughly an

“WeWork never had any revenue” False.

Gary Marcus fact-checks a claim that WeWork had no revenue, indicating this is a debunking of a false statement about the coworking company's financial performance. The post clarifies that WeWork did

“WeWork’s IPO fell apart. I’ve been calling OpenAI the (possible) WeWork of AI for a long time. I will not be surprised if their IPO plans a…

“WeWork’s IPO fell apart. I’ve been calling OpenAI the (possible) WeWork of AI for a long time. I will not be surprised if their IPO plans apart, too” - @garymarcus, June 10, two weeks before OpenAI d

When Agents Meet Electric Bus Fleet Operations: Pricing Behavior, Trade-offs, and Policy Implications in an Aggregator Framework

arXiv:2606.26400v1 Announce Type: new Abstract: Agentic systems are changing how complex operational tasks are coordinated, introducing a new paradigm for connecting heterogeneous data sources and aut

When Does Combining Language Models Help? A Co-Failure Ceiling on Routing, Voting, and Mixture-of-Agents Across 67 Frontier Models

arXiv:2606.27288v1 Announce Type: new Abstract: Multi-model LLM systems such as routing, voting, cascades, fusion, and mixture-of-agents are used to beat single-model accuracy. We show that their gain

World Action Models Enable Continual Imitation Learning with Recurrent Generative Replays

arXiv:2606.27374v1 Announce Type: cross Abstract: Going beyond predicting robot actions, World Action Models (WAMs) can also generate future visual observations. We build on this generative capability

A Zeroth-Order Deep Learning Method for Fully Nonlinear Parabolic Partial Differential Equations with Unknown Coefficients

arXiv:2606.24999v1 Announce Type: new Abstract: High-dimensional partial differential equations (PDEs) with unknown coefficients arise widely in scientific machine learning, including continuous-time

ACT-JEPA: Novel Joint-Embedding Predictive Architecture for Efficient Policy Representation Learning

arXiv:2501.14622v5 Announce Type: replace Abstract: Learning efficient representations for decision-making policies is a challenge in imitation learning (IL). Current IL methods require expert demonst

Action ControlNet: A Lightweight Delay-Aware Adapter for Smooth Asynchronous Control in Vision-Language-Action Models

arXiv:2606.25985v1 Announce Type: new Abstract: Vision-language-action (VLA) models have shown strong potential for general-purpose robot manipulation, but their inference latency remains a major obst

Adaptive Joint Compression and Synchronisation in Federated Split Learning for IoT Rainfall Prediction

arXiv:2606.25003v1 Announce Type: new Abstract: Federated split learning (FSL) enables collaborative training across bandwidth-constrained IoT devices, but repeated activation and gradient exchange cr

Adaptive Oscillatory Inductive Bias for Modeling Sharp Prosodic Dynamics in Diffusion-Based TTS

arXiv:2606.25424v1 Announce Type: cross Abstract: Diffusion-based text-to-speech (TTS) models have achieved significant improvements in speech quality. However, modeling sharp prosodic transitions and

ah, the creative S-1

ah, the creative S-1 For those unfamiliar, WeWork pioneered the creative S-1 and the subsequent failure to IPO, and Sam seems to be following a comparable path. /cc @GaryMarcus https://qz.com/1685919/

Am I More Pointwise or Pairwise? Revealing Position Bias in Rubric-Based LLM-as-a-Judge

arXiv:2602.02219v2 Announce Type: replace Abstract: Large language models are widely employed as evaluators, a paradigm commonly referred to as LLM-as-a-judge. Prior research has predominantly examine

An almost perfect prediction of the current AI predicament — and what may ultimately unravel our entire economy — from almost three years ag…

An almost perfect prediction of the current AI predicament — and what may ultimately unravel our entire economy — from almost three years ago (August 2023), at Marcus on AI: Nearly every word remains

An important point which Gary will break down on tomorrow’s @ProfGMarkets episode Tune in

An important point which Gary will break down on tomorrow’s @ProfGMarkets episode Tune in Imagine if i had a business that gave you 20 bills for 10 bills. As word spread, my customer base would grow.

Any implications for GenAI if Saylor goes down?

Gary Marcus discusses potential implications for generative AI if Michael Saylor (MicroStrategy CEO and prominent Bitcoin advocate) experiences financial or reputational difficulties. The post likely

ArteryX: A Reliable End-to-End Toolbox for Standardized Intracranial Artery Feature Extraction from 3D TOF-MRA

arXiv:2507.07920v2 Announce Type: replace-cross Abstract: Cerebrovascular research heavily relies on quantitative analysis of intracranial arteries from time-of-flight magnetic resonance angiography,

ASAP: Agent-System Co-Design for Wall-Clock-Centered Auto HPO Research for ML Experiments

arXiv:2606.25207v1 Announce Type: cross Abstract: Hyperparameter Optimization (HPO) is essential for maximizing machine learning model performance, and its core challenge is sample efficiency: finding

Benchmarking the Alignment of Data-Quality Metrics, Human Judgment and Land-Cover Segmentation Performance for Earth Observation

arXiv:2606.25128v1 Announce Type: cross Abstract: Volume and quality of datasets are crucial for deep learning model training, yet they are often constrained by availability and data acquisition costs

Beyond Next-Observation Prediction: Agent-Authored World Modeling for Sequential Decision Making

arXiv:2606.25421v1 Announce Type: new Abstract: Recent studies on world modeling for Large Language Model (LLM) agents typically formulate the learning objective as next-observation prediction. Howeve

Beyond Topology: A Morphological Symmetry Graph Representation for Locomotion Policy Learning

arXiv:2512.00727v2 Announce Type: replace Abstract: Reinforcement learning has enabled impressive locomotion skills on articulated robots, but common policy representations remain only weakly aligned

BFMTrack: Latent Sequence Optimization for Physics-Based Motion Tracking with Behavioral Foundation Models

arXiv:2606.25056v1 Announce Type: new Abstract: Behavioral Foundation Models (BFMs) offer a promising path toward universal physics-based character control by organizing a rich repertoire of physicall

Bias Fitting to Mitigate Length Bias of Reward Model in RLHF

arXiv:2505.12843v2 Announce Type: replace Abstract: Reinforcement Learning from Human Feedback (RLHF) relies on reward models to align large language models with human preferences. However, RLHF often

Causal-rCM: A Unified Teacher-Forcing and Self-Forcing Open Recipe for Autoregressive Diffusion Distillation in Streaming Video Generation and Interactive World Models

arXiv:2606.25473v1 Announce Type: new Abstract: Autoregressive video diffusion with causal diffusion transformers has emerged as a major paradigm for real-time streaming video generation and action-co

Chad Jones, economist who'll be joining Anthropic: 'In other words, with log utility it is optimal to take a 1 in 3 chance of ending human e…

Chad Jones, economist who'll be joining Anthropic: 'In other words, with log utility it is optimal to take a 1 in 3 chance of ending human existence in exchange for a 2/3 chance of dramatically raisin

safetyconnor-leahy--x

Chart of increasing revenue, pre the decline of token-maxxing, not looking at the immense costs/lack of profitability, not considering threa…

Chart of increasing revenue, pre the decline of token-maxxing, not looking at the immense costs/lack of profitability, not considering threats from open-source models, price-war dynamics etc. @noahopi

Commerge: Communication-Efficient, Robust, and Fast LiDAR Map Merging Framework for Multi-Robot Coordination in Resource-Constrained Scenarios

arXiv:2606.25386v1 Announce Type: new Abstract: By maintaining global consistency across robot teams, multi-robot LiDAR map merging enables faster exploration and efficient area coverage. However, map

Communicability-Inspired Positional Encoding (CIPE)

arXiv:2606.25293v1 Announce Type: new Abstract: Positional encodings (PEs) are essential for Transformers. Yet designing effective PEs for non-Euclidean graphs remains challenging. Such encodings shou