All entries

Post-Training Leaves Behavioral Shadows on Unrelated Decisions

arXiv:2609.29233v1 Announce Type: cross Abstract: We find that language models can transfer capabilities through task-unrelated text. Post-training typically improves language models using task-specif

Human

Content type

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

PPTBench: Can Coding Agents Reconstruct the Visual World through Structured, Editable Slides

DGX agent

arXiv:2609.29718v1 Announce Type: cross Abstract: Coding agents are beginning to act in the visual world. They now build webpages, GUIs, games, 3D scenes, diagrams, and documents. Success in such visu

local-aiarxiv-cs-ai

25 Sept 2026

Safety

Precise Convergence Speed of Clipped SGD

arXiv:2609.29458v1 Announce Type: new Abstract: We present a tightened convergence analysis of clipped gradient descent on (L_0, L_1)-smooth functions, with quantitative constants. Building on the ide

safetyarxiv-cs-lg

Predicting Emerging Topics from Outliers: A Prospective Study of Weak Signals in Embedding Space

arXiv:2609.29183v1 Announce Type: new Abstract: Some documents that embedding-based topic models initially classify as noise later become founding members of emerging topics. At publication time, howe

researcharxiv-cs-cl

Safety

Predicting Symptoms of Amotivation and Anhedonia among University Students with a Novel Oversampling Method

arXiv:2609.29690v1 Announce Type: new Abstract: University students experience disproportionately high rates of common mental health conditions, such as depression, which can impair learning, social f

safetyarxiv-cs-lg

Prefilling the Reasoning Channel: Output-Prefix Attacks on Reasoning LLMs

arXiv:2609.29775v1 Announce Type: cross Abstract: Large Language Models (LLMs) consume and produce a single sequence of text; hence, if text can be added to the beginning of the LLM's response, i.e.,

PrivDrift: Auditing User-Secret Leakage Under Topic Drift in Active LLM Conversations

arXiv:2609.30094v1 Announce Type: new Abstract: Large language models increasingly operate as persistent assistants in user-facing, shared-session, and tool-augmented settings. When users disclose sen

Progressive Skill Discovery as Access Control for Tool-Using LLM Agents: Structural Governance through Role-Scoped Capability Delivery

arXiv:2609.28693v1 Announce Type: new Abstract: Large Language Model (LLM) agents struggle to scale safely when exposed to vast enterprise toolsets. Providing an agent with access to every internal to

PROOF: Profiling Reliability of Object-Level Facts in Large Language Models

arXiv:2609.29504v1 Announce Type: cross Abstract: Aggregate factuality scores hide where a language model succeeds, which relations it confuses, and whether an answer survives innocuous changes to the

PROVE: Proof-guided Regime-aware Operator Verification for Hallucination Detection in Medical Visual Question Answering

arXiv:2609.29604v1 Announce Type: new Abstract: In medical visual question answering (VQA), hallucinations of vision-language models (VLMs) may lead to confident but incorrect responses, raising the r

researcharxiv-cs-cv

PSA for M5Ultra owners running LLMs: set your prefill step to 8192

Prefill step size affects both PP performance and your drafter fetching logits for MTP (affects dflash as well, mlx-vlm needs to be patched slightly to support chunked prefill for dflash). It also nee

local-air-localllama

PTC-Bias: Phoneme-Level Temporal Competition for Bias Retrieval and Post-Decoding Correction in Speech LLMs

arXiv:2609.28727v1 Announce Type: new Abstract: Contextual biasing improves rare-word recognition in speech large language models (SpeechLLMs), but efficiently exploiting large bias lists remains chal

local-aiarxiv-cs-cl

PUBG Ally: A Conversational Embodied Agent as an AI Teammate

arXiv:2609.29837v1 Announce Type: new Abstract: We introduce PUBG Ally, an embodied agent for PUBG: BATTLEGROUNDS that can reason, act autonomously, and play alongside players as a voice-enabled teamm

local-aiarxiv-cs-ai

pylazaro: a Python package for anglicism extraction in Spanish

arXiv:2609.29276v1 Announce Type: new Abstract: Lexical borrowings are words from one language that are introduced into another language. Identifying lexical borrowings in text is a relevant task for

researcharxiv-cs-cl

QINA: Quantum-Inspired Nonlinear Adapters for Pretrained Vision Models

arXiv:2609.29592v1 Announce Type: cross Abstract: Adapting large pretrained vision models under limited data and frozen-backbone constraints remains a central challenge in transfer learning. While lig

Quantum Attention by Overlap Interference: Predicting Classical and Many-Body Quantum Sequences

arXiv:2602.06699v2 Announce Type: replace-cross Abstract: We propose a variational quantum implementation of self-attention (QSA)-the core operation in transformers and large language models-which pre

researcharxiv-cs-cl

Agents

Quoting John Gruber

Muse is getting a lot of attention — including mine — because it’s both groundbreaking technically (each user gets their own entire persistent Linux VM running in Meta’s cloud) and because it’s packag

agentssimon-willison

Qwen-Planner-Agent: A Closed-Loop AI-for-AI Framework for Real-World Mobile Planner Agents

arXiv:2609.29892v1 Announce Type: new Abstract: The rapid progression of large language models is extending AI from passive content generation into the active workflows of engineering and scientific d

Qwen3.8-27B: Using KV Cache Transplants to Boost Output Quality

model-releasesr-localllama

Since my last post, I've been thinking about different options for dynamic performance degradation, trying to squeeze as much high-quality inference out of my GPU as I can. Over the weekend I read thi

Qwengram-0.8B: I transferred Qwen3.8 Flash-Next’s n-gram memory into Qwen3.5-0.8B — 5.05% lower validation perplexity

model-releasesr-localllama

I’ve been experimenting with whether Qwen3.8-Flash-Next’s pretrained PLE n-gram memory can improve a much smaller Qwen3.5-0.8B model. I trained the 0.8B setup with limited resources, mostly using free

R-DEIM Net: An Efficient Rationale-Augmented Dual-Expert Interaction Model for Paraphrase Detection

arXiv:2609.30100v1 Announce Type: cross Abstract: Recent advances in paraphrase detection reveal a fundamental trade-off: large language models achieve high accuracy but require high computation, whil

RACaP: Agentic Reasoning, Acting, and Coding as Policies for Evolvable Robot Learning

model-releasesarxiv-cs-ro

arXiv:2609.29394v1 Announce Type: new Abstract: General-purpose robot agents must learn from experience, transfer to new tasks, and act efficiently. Code as Policies (CaP) methods generate and repair

Agents

RADAR: Readiness for AI Discovery and Agentic Reach

arXiv:2609.28480v1 Announce Type: cross Abstract: Governments increasingly meet citizens through an AI system rather than a website. RADAR (Readiness for AI Discovery and Agentic Reach) measures wheth

agentsarxiv-cs-lg

Industry

Radical Ventures: AI neolabs, many lacking products, markets, or revenue, raised $24B in the past two quarters, nearly 5x OpenAI and Anthropic pre-ChatGPT (Louise Lucas/Financial Times)

Louise Lucas / Financial Times: Radical Ventures: AI neolabs, many lacking products, markets, or revenue, raised $24B in the past two quarters, nearly 5x OpenAI and Anthropic pre-ChatGPT — It is true

industrytechmeme

RAPID: Robot Agentic Programming from Demonstrations

arXiv:2609.30249v1 Announce Type: cross Abstract: Coding agents have demonstrated enormous success in solving complex programming problems. To leverage their potential for robot systems, this work int

Tutorials

RAPTOR: RAndom-projection Physics-informed Transient sOlveR

arXiv:2609.29714v1 Announce Type: cross Abstract: The complexity of time-domain simulation of modern power systems has increased significantly because converter-based resources introduce control dynam

tutorialsarxiv-cs-lg

RD-JEPA: Predictive latent pretraining for few-trajectory transfer across reaction--diffusion equations

arXiv:2609.29403v1 Announce Type: new Abstract: Learning surrogates for time-dependent partial differential equations often requires a new simulation corpus when the governing operator changes. We int

Reachability-Based Formal Verification of Graph Neural Networks with Node and Edge Features

arXiv:2609.30079v1 Announce Type: cross Abstract: Graph neural networks (GNNs) have become a prominent approach for developing fast, topology-aware surrogates in electric power systems, supporting tas

Applications

Real-Time Force Regulation for Whole-Hand Dexterous Grasping

arXiv:2609.30082v1 Announce Type: new Abstract: Robust dexterous grasping requires maintaining physical stability despite contacts interactively evolving across the entire hand. A precomputed force di

applicationsarxiv-cs-ro

Reasoning Instructions Can Break Answer Decoding in Vision--Language Models

arXiv:2609.29278v1 Announce Type: cross Abstract: Chain-of-thought (CoT) instructions can distort multiple-choice VLM evaluation when a scorer appends a reasoning cue but reads answer-label logits bef

ReCalMatch:Reliability-Calibrated Semantic Guidance for Semi-Supervised Fine-Grained Recognition

arXiv:2609.29678v1 Announce Type: cross Abstract: Semi-supervised fine-grained visual recognition is highly vulnerable to overconfident pseudo-label errors: visually similar categories frequently prod

RECLAIM: Can Agents Reproduce the Claims of Machine Learning Papers?

arXiv:2609.28850v1 Announce Type: new Abstract: Reproducing a machine learning paper involves most research steps, from installing software and debugging to running experiments, work that AI agents in

Recommended benchmark. I expect voice to become one of the main ways people interact with robots and physical AI. That makes speech recognit…

Recommended benchmark. I expect voice to become one of the main ways people interact with robots and physical AI. That makes speech recognition in real conversations much more important than it seems

model-releasesdair-ai--x

Reconstructing short-lived particles using hypergraph representation learning

model-releasesarxiv-cs-lg

arXiv:2402.10149v3 Announce Type: cross Abstract: In collider experiments, the kinematic reconstruction of heavy, short-lived particles is vital for precision tests of the Standard Model and in search

Recoverable Geographic Location Information in Earth-Observation Embeddings

arXiv:2609.29151v1 Announce Type: new Abstract: Earth-observation (EO) foundation models provide reusable embeddings, yet downstream task accuracy does not reveal whether these representations encode

researcharxiv-cs-cv

Refusing Everything Looks Safe: Restoring the Benign Arm to Encoded-Prompt Evaluation

arXiv:2609.26176v2 Announce Type: replace-cross Abstract: Encoded-prompt attacks are evaluated almost entirely on their harmful arm: a benchmark sends obfuscated harmful requests and reports how often

Reinforcement Learning with Verifiable Rewards for Small Search Agents

arXiv:2609.28765v1 Announce Type: new Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) performs well on problems with clear rewards, such as mathematics and coding, but whether it also

Tools

Reliable, fast inference is essential. As Ted shared at Apsara Conference 2026, Together makes it easy for teams to get access to frontier o…

Reliable, fast inference is essential. As Ted shared at Apsara Conference 2026, Together makes it easy for teams to get access to frontier open models without needing to manage any of the infrastructu

toolstogether-ai--x

Remember October 2022 when you doused Galactica with vitriol? Galactica was a 120b-parameter LLM-based system from Meta-FAIR designed to hel…

model-releasesyann-lecun--x

Remember October 2022 when you doused Galactica with vitriol? Galactica was a 120b-parameter LLM-based system from Meta-FAIR designed to help scientist write papers. It was open sourced (link below).

Representation World Model: Learning States, Transition and Executable Plans in Representation

arXiv:2609.29171v1 Announce Type: cross Abstract: We propose the Representation World Model (RWM), which learns states, transitions, and executable plans directly in representation space. Unlike exist

local-aiarxiv-cs-cv

Requirement-Bound Verified Commissioning: A Frozen Four-Billion-Parameter Local Model as a Candidate Generator under an External Acceptance Layer with Verification and Release Authority

arXiv:2609.30219v1 Announce Type: cross Abstract: An acceptance protocol is developed for sensor-coordinate and polarity binding in mechatronic commissioning. Candidate generation is separated from re

Safety

Res-HIL: Human-Guided Residual Reinforcement Learning for Sample-Efficient Dexterous Manipulation

arXiv:2609.30023v1 Announce Type: new Abstract: Imitation learning enables robots to acquire manipulation skills from demonstrations, but the resulting policies can fail outside the training data, whi

safetyarxiv-cs-ro

Industry

Researchers add details to the Hugging Face incident, including OpenAI agents creating ~1M shortened URLs to encode information in an attempt to solve CAPTCHAs (Dylan Freedman/New York Times)

Dylan Freedman / New York Times: Researchers add details to the Hugging Face incident, including OpenAI agents creating ~1M shortened URLs to encode information in an attempt to solve CAPTCHAs — A new

industrytechmeme

Residual Correlation as a Diagnostic for Joint-Uncertainty Gains from GP Coregionalisation

arXiv:2609.30085v1 Announce Type: new Abstract: In multi-target regression, correlated targets are often coupled through multi-output Gaussian processes with an intrinsic model of coregionalisation (G

researcharxiv-cs-lg

Response-state Learning for Transferable Vibrational Spectroscopic Characterization with Electron Prior

arXiv:2609.28935v1 Announce Type: new Abstract: Vibrational spectral prediction can become inaccurate when localized stereoelectronic environments perturb intermediate response states and high-risk re

researcharxiv-cs-lg

Retrieve-to-Localize: Bridging Large Language Models and LiDAR Geometry for Spatial Grounding

model-releasesarxiv-cs-cv

arXiv:2609.29835v1 Announce Type: new Abstract: LiDAR provides precise geometric information for spatial perception tasks such as object detection in autonomous driving and outdoor robotics. However,

Return or Revise? Learning When Revision Helps Retrieval-Augmented QA

model-releasesarxiv-cs-cl

arXiv:2609.30087v1 Announce Type: new Abstract: We consider the decision of whether to return an existing draft answer or revise it using retrieved evidence, as in answer-revision systems. Draft confi

Revalidation Beats Stateful Routing for Scientific Surrogates Under Distribution Shift

arXiv:2609.29715v1 Announce Type: cross Abstract: Surrogate models are often chosen during development and then left in place as new measurements arrive. That practice becomes risky when noise, input