All entries

The Geometric Structure of Models Learning Sparse Data

arXiv:2605.08464v1 Announce Type: new Abstract: The manifold hypothesis (MH) is often used to explain how machine learning can overcome the curse of dimensionality. However, the MH is only applicable

safetyarxiv-cs-lg

The Geometric Wall: Manifold Structure Predicts Layerwise Sparse Autoencoder Scaling Laws

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

DGX agent

arXiv:2605.09887v1 Announce Type: cross Abstract: Sparse autoencoders (SAEs) operationalise the linear representation hypothesis: they reconstruct model activations as sparse linear combinations of in

model-releasesarxiv-cs-ai

12 May 2026

The Geometry of Forgetting: Temporal Knowledge Drift as an Independent Axis in LLM Representations

arXiv:2605.09195v1 Announce Type: new Abstract: Large language models confidently produce outdated answers, and no existing method can detect them. We show this is not an engineering failure but a str

The Global Empirical NTK: Self-Referential Bias and Dimensionality of Gradient Descent Learning

model-releasesarxiv-cs-lg

arXiv:2605.08746v1 Announce Type: new Abstract: In training a neural network with gradient descent (GD), each iteration induces a linear operator that governs first-order updates to a model's internal

The Gordian Knot for VLMs: Diagrammatic Knot Reasoning as a Hard Benchmark

arXiv:2605.09900v1 Announce Type: new Abstract: A vision-language model can look at a knot diagram and report what it sees, yet fail to act on that structure. KnotBench pairs an 858,318-image corpus f

The Grounding Gap: How LLMs Anchor the Meaning of Abstract Concepts Differently from Humans

arXiv:2605.08837v1 Announce Type: cross Abstract: Abstract concepts - justice, theory, availability - have no single perceivable referent; in the human brain, their meaning emerges from a web of exper

The Impact of Editorial Intervention on Detecting Native Language Traces

arXiv:2605.10216v1 Announce Type: new Abstract: Native Language Identification (NLI) is the task of determining an author's native language (L1) from their non-native writings. With the advent of huma

researcharxiv-cs-cl

The Invisible Handshake: Persistent Overpricing by Adaptive Market Agents

arXiv:2510.15995v3 Announce Type: replace-cross Abstract: We study overpricing in a repeated game between two representative agents: a market maker, who controls market liquidity, and a market taker,

The Last Word Often Wins: A Format Confound in Chain-of-Thought Corruption Studies

arXiv:2605.10799v1 Announce Type: cross Abstract: Corruption studies, the primary tool for evaluating chain-of-thought (CoT) faithfulness, identify which chain positions are 'computationally important

The Metacognitive Probe: Five Behavioural Calibration Diagnostics for LLMs

arXiv:2605.09844v1 Announce Type: new Abstract: The Metacognitive Probe is an exploratory five-task, 15-slot diagnostic that decomposes an LLM's confidence behaviour into five behaviourally-distinct d

the Mini Shai-Hulud attack is scary because it attacks new AI coding workflows like CI, editor hooks, agent configs, etc

The Mini Shai-Hulud attack targets emerging AI-assisted development workflows by compromising multiple integration points including continuous integration systems, code editor hooks, and AI agent conf

agentsyohei-nakajima--x

The Mixing method: low-rank coordinate descent for semidefinite programming with diagonal constraints

arXiv:1706.00476v4 Announce Type: replace-cross Abstract: In this paper, we propose a low-rank coordinate descent approach to structured semidefinite programming with diagonal constraints. The approac

Tutorials

The new Cofounder site is literally a step-by-step guide on how to start a company. When i started my first company there was all sorts of s…

tutorialsyohei-nakajima--x

The new Cofounder site is literally a step-by-step guide on how to start a company. When i started my first company there was all sorts of stuff i had to learn and all of the guides were SEO maxxed sl

The newest AI boom pitch: Host a mini data center at your home

Span, Nvidia, and PulteGroup are testing home-based data center nodes that use unused grid capacity to cut costs, ease community pushback, and support growing AI demand. The Span installation bundles

industryars-technica

The Observable Wasserstein Distance

arXiv:2605.09916v1 Announce Type: cross Abstract: We introduce the observable Wasserstein distance, a framework for deriving lower bounds on the Wasserstein distance between probability measures on Po

The Open-Box Fallacy: Why AI Deployment Needs a Calibrated Verification Regime

arXiv:2605.10601v1 Announce Type: new Abstract: AI deployment in sensitive domains such as health care, credit, employment, and criminal justice is often treated as unsafe to authorize until model int

researcharxiv-cs-ai

The Pokemon Theorem and other Fairness Impossibility Results

arXiv:2605.09221v1 Announce Type: cross Abstract: Fairness impossibility results often look like distinct scalar incompatibility statements. We show that several share one RKHS geometry: fairness crit

The Polynomial Counting Capabilities of Message Passing Neural Networks

arXiv:2605.10393v1 Announce Type: new Abstract: The counting power of Message Passing Neural Networks (MPNN) has been the subject of many recent papers, showing that they can express logic that involv

The Power of Second Order Methods for Sequence Preconditioning

arXiv:2605.08390v1 Announce Type: new Abstract: Sequence prediction methods for dynamical systems with long memory, i.e. marginally stable systems, typically achieve regret that grows polynomially wit

The Procrustean Bed of Time Series: The Optimization Bias in Point-wise Loss Functions

arXiv:2512.18610v3 Announce Type: replace Abstract: Intuitively, a more deterministic time series should be easier to forecast. However, point-wise loss functions (e.g., MSE and MAE), serving as diffe

safetyarxiv-cs-lg

The Propagation Field: A Geometric Substrate Theory of Deep Learning

arXiv:2605.08529v1 Announce Type: new Abstract: Modern deep learning treats neural networks primarily as endpoint functions from inputs to outputs. Inspired by the shift from force to geometry in phys

The Realignment Problem: When Right becomes Wrong in LLMs

model-releasesarxiv-cs-cl

arXiv:2511.02623v2 Announce Type: replace Abstract: Post-training alignment of large language models (LLMs) relies on large-scale human annotations guided by policy specifications that change over tim

The Reciprocity Gradient

arXiv:2605.08323v1 Announce Type: cross Abstract: Communication is fundamental to sustaining reciprocity and cooperation in strategic interactions. We identify and formulate the influence attribution

agentsarxiv-cs-ai

The Rise of Sports Intelligence: How the Lakehouse Turns Tracking Data into Competitive Advantage

Databricks explores how lakehouse architecture enables sports organizations to process and analyze player tracking data at scale, converting raw motion capture and sensor information into actionable c

industrydatabricks

The Safety-Aware Denoiser for Text Diffusion Models

arXiv:2605.08116v1 Announce Type: cross Abstract: Recent work on text diffusion models offers a promising alternative to autoregressive generation, but controlling their safety remains underexplored.

The Sample Complexity of Uniform Approximation for Multi-Dimensional CDFs and Fixed-Price Mechanisms

arXiv:2602.10868v2 Announce Type: replace Abstract: We study the sample complexity of learning a uniform approximation of an n-dimensional cumulative distribution function (CDF) within an error epsilo

The scale of the infra on HF is insane. If you're still hosting models, datasets, agent memory,... in S3 or R2, talk to use and we can help …

Hugging Face offers substantial infrastructure capabilities for hosting machine learning models, datasets, and agent memory systems. The statement suggests that organizations currently using alternati

agentsclem-delangue--x

The silent removal of Study Mode from ChatGPT is a big mistake (both Claude and Gemini still have theirs) We have enough evidence that using…

model-releasesethan-mollick--x

The silent removal of Study Mode from ChatGPT is a big mistake (both Claude and Gemini still have theirs) We have enough evidence that using AI in assistant mode to study can hurt learning because it

The Silent Vote: Improving Zero-Shot LLM Reliability by Aggregating Semantic Neighborhoods

arXiv:2605.09739v1 Announce Type: cross Abstract: Large Language Models are increasingly used as zero-shot classifiers in complex reasoning tasks. However, standard constrained decoding suffers from a

Tools

The talks that will define what comes next in creativity. Vibecon brings the voices shaping code-as-medium to the stage in NYC June 17–18. G…

Vibecon is a conference hosted by Replit taking place in New York City on June 17-18 that features speakers and discussions about code-as-medium and the future of creative coding. The event brings tog

toolsreplit--x

The Trap of Trajectory: Towards Understanding and Mitigating Spurious Correlations in Agentic Memory

arXiv:2605.09330v1 Announce Type: cross Abstract: Agentic memory enables LLMs to persist information beyond a single context window and reuse it in later decisions, but it also introduces a new vulner

Local Ai

The Truth Lies Somewhere in the Middle (of the Generated Tokens)

arXiv:2605.09969v1 Announce Type: cross Abstract: How should hidden states generated autoregressively be collapsed into a representation that reflects a language model's internal state? Despite tokens

local-aiarxiv-cs-cl

Tutorials

The two clocks and the innovation window: When and how generative models learn rules

arXiv:2605.10019v1 Announce Type: cross Abstract: Generative models trained on finite data face a fundamental tension: their score-matching or next-token objective converges to the empirical training

tutorialsarxiv-cs-ai

Applications

The US' Centers for Medicare & Medicaid Services is testing ACCESS, an outcome-based payment model for AI-driven medical care, with 150 tech companies (Connie Loizos/TechCrunch)

Connie Loizos / TechCrunch: The US' Centers for Medicare & Medicaid Services is testing ACCESS, an outcome-based payment model for AI-driven medical care, with 150 tech companies — Neil Batlivala has

applicationstechmeme

The US DOD says it is deploying Mythos to find and patch software vulnerabilities across the US government, even as it works on a transition away from Anthropic (Reuters)

Reuters: The US DOD says it is deploying Mythos to find and patch software vulnerabilities across the US government, even as it works on a transition away from Anthropic — WASHINGTON, May 12 (Reuters)

industrytechmeme

The US FCC approves EchoStar's sale of approximately 65MHz of spectrum to SpaceX and 50MHz to AT&T (Christian Martinez/Reuters)

Christian Martinez / Reuters: The US FCC approves EchoStar's sale of approximately 65MHz of spectrum to SpaceX and 50MHz to AT&T — The U.S. Federal Communications Commission's Wireless Telecommunicati

industrytechmeme

The US House Oversight Committee launches a probe into potential conflicts in Sam Altman's personal investments; letter: several GOP AGs call for an SEC review (Wall Street Journal)

Wall Street Journal: The US House Oversight Committee launches a probe into potential conflicts in Sam Altman's personal investments; letter: several GOP AGs call for an SEC review — Republican-led Ho

model-releasestechmeme

The Value of Mechanistic Priors in Sequential Decision Making

arXiv:2605.10018v1 Announce Type: new Abstract: Hybrid mechanistic models, physical priors with learned residuals, promise to reduce the data required for good decisions, but have no computable criter

safetyarxiv-cs-lg

The Wittgensteinian Representation Hypothesis: Is Language the Attractor of Multimodal Convergence?

arXiv:2605.09352v1 Announce Type: new Abstract: Understanding why independently trained neural networks from different modalities converge toward shared representations, and where this convergence lea

The World is Not Mono: Enabling Spatial Understanding in Large Audio-Language Models

arXiv:2601.02954v3 Announce Type: replace-cross Abstract: Large audio-language models have made rapid progress in recognizing what is present in an audio clip, but spatial audio-language understanding

The Wristband Gaussian Loss: Deterministic, Composable Latents via a Sphere-Interval Decomposition

model-releasesarxiv-cs-lg

arXiv:2605.08749v1 Announce Type: new Abstract: We present the Wristband Gaussian Loss, a deterministic batch loss for Gaussianizing point embeddings without sampling, KL terms, or iterative transport

There will be no AI jobpocalypse. The story that AI will lead to massive unemployment is stoking unnecessary fear. AI — like any other techn…

There will be no AI jobpocalypse. The story that AI will lead to massive unemployment is stoking unnecessary fear. AI — like any other technology — does affect jobs, but telling overblown stories of l

safetyandrew-ng--x

Thermal-Det: Language-Guided Cross-Modal Distillation for Open-Vocabulary Thermal Object Detection

arXiv:2605.10130v1 Announce Type: new Abstract: Existing open-vocabulary detectors focus on RGB images and fail to generalize to thermal imagery, where low texture and emissivity variations challenge

safetyarxiv-cs-cv

Think as Needed: Geometry-Driven Adaptive Perception for Autonomous Driving

arXiv:2605.10117v1 Announce Type: cross Abstract: Autonomous driving scenes range from empty highways to dense intersections with dozens of interacting road users, yet current 3D detection models appl

agentsarxiv-cs-ai

Thinking Machines drops a new, highly responsive model designed for humanlike interactions in real time

Thinking Machines Lab Inc., the artificial intelligence research startup founded by former OpenAI Group PBC Chief Technology Officer Mira Murati, wants to move beyond the era of “turn-based” AI intera

industrysiliconangle

Thinking with Novel Views: A Systematic Analysis of Generative-Augmented Spatial Intelligence

arXiv:2605.10588v1 Announce Type: new Abstract: Current Large Multimodal Models (LMMs) struggle with spatial reasoning tasks requiring viewpoint-dependent understanding, largely because they are confi

researcharxiv-cs-cv

This NVIDIA remains the strongest platform for large-model inference at scale. Prefill/decode disaggregation, Blackwell-native quantization,…

model-releasesperplexity--x

This NVIDIA remains the strongest platform for large-model inference at scale. Prefill/decode disaggregation, Blackwell-native quantization, custom kernels, and rack-scale NVLink turn GB200 into faste

This was posted exactly 3 years ago. What 3 years of progress look like is wild. Can't wait for the next 3 years.

industrycristobal-valenzuela--x

This was posted exactly 3 years ago. What 3 years of progress look like is wild. Can't wait for the next 3 years. guess who has access to runway #gen2 now? 👀 first quick trial for animated #aicinema,