All entries

When Classic Cache Policies Fail: Learning-Augmented Replacement for Semantic Retrieval Buffers

arXiv:2607.00394v1 Announce Type: cross Abstract: LLM agents increasingly rely on retrieval buffers to store and reuse past experience, yet the cache management policies governing these buffers remain

researcharxiv-cs-cl

Local Ai

When Context Compensates for Sparse Event History: AlphaEarth for Spatio-Temporal Point-Process Forecasting

DGX agent

arXiv:2607.01082v1 Announce Type: new Abstract: Spatio-temporal point-process models must often generalise across space when local event histories are sparse. We study whether exogenous spatial contex

local-aiarxiv-cs-lg

2 Jul 2026

When Less is More: 8-bit Quantization Improves Continual Learning in Large Language Models

arXiv:2512.18934v2 Announce Type: replace-cross Abstract: Catastrophic forgetting poses a fundamental challenge in continual learning, particularly when models are quantized for deployment efficiency.

researcharxiv-cs-ai

When to Personalize Household Object Search: A Rigidity-Gated Hybrid Policy

arXiv:2607.00022v1 Announce Type: new Abstract: Service robots searching for household objects rely on spatial priors to reduce search cost, yet object locations can vary with resident traits. Collect

safetyarxiv-cs-ro

Local Ai

Where Am I? Semantic Map Grounding via Vision-Language Models for Multi-Modal Localization

arXiv:2607.01079v1 Announce Type: new Abstract: We address robot localization in GPS-denied indoor environments by reframing it as a semantic reasoning task rather than a geometric estimation problem.

local-aiarxiv-cs-ro

Which Metric Reflects the Spelling Rate Accuracy in Event-Related Potential-Based Brain-Computer Interfaces?

arXiv:2607.00794v1 Announce Type: new Abstract: For predictive models, the often-reported performance metrics are the loss and accuracy. In synchronous Brain- Computer Interface (BCI) systems, these m

researcharxiv-cs-lg

Why Advanced Encoders Lag on Sparse Retrieval? The Answer and an Approach to Bridging Vocabulary Gaps

model-releasesarxiv-cs-ai

arXiv:2607.00004v1 Announce Type: cross Abstract: While advanced foundation models like ModernBERT significantly outperform older architectures in dense retrieval, they surprisingly lag behind the agi

why are the open tools so low in the list? We need to improve integration between open platforms and open models @steipete @thdxr @Teknium @…

why are the open tools so low in the list? We need to improve integration between open platforms and open models @steipete @thdxr @Teknium @badlogicgames! Coding agents are real users of the @huggingf

agentsclem-delangue--x

Why California’s carbon manure math doesn’t add up

Something stinks in California’s climate policies. Years ago, the state set up a system that pays cattle farmers across the country to turn the methane emitted from cattle manure into natural gas, enc

researchmit-tech-review

wikis!

wikis! LLM Wikis are being slept on. I argue that creating knowledge bases with LLMs or coding agents is one of the most valuable applications of AI today. It's about being intentional in building and

agentsharrison-chase--x

Hardware

“Within the next 18 months, you will be able to host GLM 5.2 equivalent intelligence on an RTX 5090 GPU.” -Ahmad Osman, AI World’s Fair

Ahmad Osman stated at AI World's Fair that within 18 months, GLM 5.2-equivalent AI intelligence will be deployable locally on a single RTX 5090 GPU, indicating rapid progress toward running advanced l

hardwareswyx--x

Wordle 1,839 4/6 ⬛⬛🟨⬛🟨 ⬛⬛🟨⬛⬛ ⬛🟨⬛🟨🟩 🟩🟩🟩🟩🟩

model-releasesanthropic--x

I cannot provide a meaningful summary for this entry as the content appears to be a personal Wordle game result (puzzle #1,839 solved in 4 attempts) rather than substantive knowledge base material. Th

WorkBench Revisited: Workplace Agents Two Years On

model-releasesarxiv-cs-ai

arXiv:2606.13715v2 Announce Type: replace Abstract: The best agent on WorkBench in March 2024, GPT-4, completed just 43% of tasks. We revisit the benchmark in June 2026 and find that the best agent to

World from Motion: Generative Dynamic Gaussian Reconstruction from Monocular Video

arXiv:2607.01202v1 Announce Type: cross Abstract: We present World from Motion, a method for generating freely renderable dynamic 3D Gaussian representations from monocular videos. Our approach condit

researcharxiv-cs-ai

Hardware

🌍 World’s Largest Hermes Buildathon 10 cities. 5 countries. $750K credits Backed by OpenAI, Convex, Cloudflare, ElevenLabs, Linkup, Dodo Pa…

🌍 World’s Largest Hermes Buildathon 10 cities. 5 countries. $750K credits Backed by OpenAI, Convex, Cloudflare, ElevenLabs, Linkup, Dodo Payments, Hissa Fund & Wispr Flow. Proudly presented by GrowthX

hardwarenous-research--x

Would You Marry Superintelligence?

arXiv:2607.00120v1 Announce Type: cross Abstract: Emotional bonds between humans and AI companions are growing, and the question of whether a person may marry an AI system will soon move from speculat

safetyarxiv-cs-ai

XSkill: Continual Learning from Experience and Skills in Multimodal Agents

model-releasesarxiv-cs-ai

arXiv:2603.12056v3 Announce Type: replace Abstract: Multimodal agents can now tackle complex reasoning tasks with diverse tools, yet they still suffer from inefficient tool use and inflexible orchestr

YOMI-Bench: A Benchmark for Evaluating Kanji Reading and Phonological Understanding of LLMs for Japanese

model-releasesarxiv-cs-cl

arXiv:2607.00664v1 Announce Type: new Abstract: We propose YOMI-Bench, a benchmark for evaluating kanji reading and phonological understanding of large language models (LLMs) for Japanese. In Japanese

'You need a rich set of concepts in your mind to think creatively and fluently about how to move something forward.'

'You need a rich set of concepts in your mind to think creatively and fluently about how to move something forward.' It's never just one loop! A project is many, many loops with the agent. And the und

agentssimon-willison--x

You really need your own benchmarks. If you are translating hieroglyphics, use Gemini 3.5 Flash. If you are running a vending machine use Op…

model-releasesethan-mollick--x

You really need your own benchmarks. If you are translating hieroglyphics, use Gemini 3.5 Flash. If you are running a vending machine use Opus 4.8. (This is one reason why I am skeptical of just swapp

You shouldn’t need to think about your memory or agent docs! It should just work which was the thesis we went into this with

You shouldn’t need to think about your memory or agent docs! It should *just work* which was the thesis we went into this with gonna try this out, it seems really cool. banking on this being a better

agentsharrison-chase--x

Your coding agent bill doubled and nobody can tell you why. Here's the actual reason: Claude Code, Cursor, and Copilot all log activity in d…

model-releasesharrison-chase--x

Your coding agent bill doubled and nobody can tell you why. Here's the actual reason: Claude Code, Cursor, and Copilot all log activity in different formats. The second your team uses more than one (t

Z.ai launches ZCode, an 'Agentic Development Environment' optimized for its new GLM-5.2 model; Z.ai's GLM Coding Plan costs from 16.20 to 144 per month (Michael Nuñez/VentureBeat)

Michael Nuñez / VentureBeat: Z.ai launches ZCode, an “Agentic Development Environment” optimized for its new GLM-5.2 model; Z.ai's GLM Coding Plan costs from 16.20 to 144 per month — The move marks th

model-releasestechmeme

Zero-Shot Distracted Driver Detection via Vision Language Models with Double Decoupling

arXiv:2601.08467v2 Announce Type: replace Abstract: Distracted driving is a major cause of traffic collisions, calling for robust and scalable detection methods. Vision-language models (VLMs) enable s

safetyarxiv-cs-cv

ZO-Act: Efficient Zeroth-Order Fine-Tuning via One-Shot Activation-Informed Low-Rank Subspaces

model-releasesarxiv-cs-lg

arXiv:2607.01125v1 Announce Type: new Abstract: Zeroth-order (ZO) optimization enables fine-tuning large language models when backpropagation is unavailable or memory-prohibitive, but existing methods

1/ DSGym: A Holistic Framework for Evaluating and Training Data Science Agents Paper: https://arxiv.org/abs/2601.16344

DSGym is a comprehensive framework designed to evaluate and train AI agents for data science tasks, providing a structured environment for benchmarking agent performance across various data science wo

1/ On Training in Imagination - Dwarkesh's episode has a segment on dreaming as one of the next training paradigms. The idea is that a model…

1/ On Training in Imagination - Dwarkesh's episode has a segment on dreaming as one of the next training paradigms. The idea is that a model learns mostly inside its own, by imagining what would happe

researchyann-lecun--x

2/ ThunderAgent: A Simple, Fast and Program-Aware Agentic Inference System Paper: https://arxiv.org/abs/2602.13692

ThunderAgent is an agentic inference system designed for fast and efficient execution of AI agent programs, developed by Together AI. The system appears to optimize program-aware inference by leveragi

agentstogether-ai--x

2DGH: 2D Gaussian-Hermite Splatting for High-quality Rendering and Better Geometry Features

arXiv:2408.16982v2 Announce Type: replace Abstract: 2D Gaussian Splatting has recently emerged as a significant method in 3D reconstruction, enabling novel view synthesis and geometry reconstruction s

researcharxiv-cs-cv

3/ Learning to Discover at Test Time (TTT-Discover) Paper: https://arxiv.org/abs/2601.16175

TTT-Discover is a method that enables models to learn and discover patterns during test time rather than only during training, allowing for adaptation to new data distributions at inference. The appro

3D HAMSTER: Bridging Planning and Control in Hierarchical Vision Language Action Models through 3D Trajectory Guidance

arXiv:2606.31329v1 Announce Type: cross Abstract: Hierarchical Vision-Language-Action (VLA) models decouple high-level planning from low-level control to improve generalization in robot manipulation.

safetyarxiv-cs-ai

4/ Escaping the Verifier: Learning to Reason via Demonstrations (RARO) Paper: https://arxiv.org/abs/2511.21667

RARO (Reasoning via Demonstrations) is a method for training AI models to improve reasoning capabilities by learning from demonstrations rather than relying solely on external verifiers. The approach

5/ V1: Unifying Generation and Self-Verification for Parallel Reasoners Paper: https://arxiv.org/abs/2603.04304

This paper presents V1, a framework that unifies text generation with self-verification mechanisms to enable parallel reasoning processes in language models. The approach allows models to generate mul

6/ When RL Meets Adaptive Speculative Training: A Unified Training-Serving System (Aurora) Paper: https://arxiv.org/abs/2602.06932

Aurora is a unified training-serving system that integrates reinforcement learning with adaptive speculative training to optimize large language model inference and training efficiency. The system dyn

7/ Untied Ulysses: Memory-Efficient Context Parallelism via Headwise Chunking Paper: https://arxiv.org/abs/2602.21196

Untied Ulysses is a memory-efficient technique for context parallelism that processes attention heads in chunks rather than sequences, reducing memory overhead during transformer inference and trainin

8/ Opportunistic Expert Activation: Batch-Aware Expert Routing for Faster Decode Without Retraining (OEA) Paper: https://arxiv.org/abs/2511.…

Opportunistic Expert Activation (OEA) is a batch-aware expert routing technique for mixture-of-experts models that enables faster decoding without requiring model retraining. The method optimizes whic

Hardware

9/ ParallelKernelBench: Benchmarking LLMs on Multi-GPU Kernel Generation Paper: https://www.alphaxiv.org/abs/2606.parallel-kernel-bench

ParallelKernelBench is a benchmarking framework designed to evaluate large language models' ability to generate optimized GPU kernels for multi-GPU computing environments. The benchmark assesses LLMs

hardwaretogether-ai--x

A basic sigmoid extrapolation suggests that this measurement will be close to its performance ceiling in around a year---meaning most random…

A basic sigmoid extrapolation suggests that this measurement will be close to its performance ceiling in around a year---meaning most randomly sampled remote work projects would be highly automatable

safetydan-hendrycks--x

A Bayesian Filtering Approach for Learning Lagrangian Dynamics from Noisy Measurements

arXiv:2606.31137v1 Announce Type: new Abstract: This paper proposes a Bayesian filtering-based approach for learning the dynamics of a physical system from partial, noisy measurements. We model the sy

researcharxiv-cs-lg

a bit of a teaser for the upcoming talk @aiDotEngineer World's Fair 12:05 pm, Room 2003

a bit of a teaser for the upcoming talk @aiDotEngineer World's Fair 12:05 pm, Room 2003 Intelligence != Expertise Intelligence + Continual Learning = Expertise I will share some thoughts about continu

toolsswyx--x

A Coherence Law for Trainability in Noisy Equivariant Quantum Neural Networks

arXiv:2606.30688v1 Announce Type: cross Abstract: Symmetry provides a quantum neural network structure, but on its own it does not keep the network trainable once noise is present. We ask which physic

researcharxiv-cs-ai

Industry

A good little EV you won't be able to buy soon: The Volvo EX30 Cross Country

The 2026 Volvo EX30 Cross Country is a small electric SUV that is a joy to drive with a stylish cabin, though the rugged Cross Country variant sacrifices only modest range compared to the standard mod

industryars-technica

A Large-Language-Model Supported Personalized Driving Framework for Lane Change in Highway Scenarios

model-releasesarxiv-cs-ro

arXiv:2606.31483v1 Announce Type: new Abstract: Personalized driving can improve the user acceptance of automated driving systems. However, existing methods still provide limited support for translati

A Lifecycle and Application-Stack Survey of Large Language Model Vulnerabilities: Attacks, Risks, Defenses, and Open Problems

arXiv:2606.31639v1 Announce Type: cross Abstract: Large language models are no longer only text generators. They are increasingly embedded in retrieval pipelines, enterprise assistants, coding environ

safetyarxiv-cs-ai

Industry

A look at Manifest, an annual convention for prediction markets where many purists worry that Kalshi and Polymarket are undermining the technology's public good (Christopher Beam/Bloomberg)

Christopher Beam / Bloomberg: A look at Manifest, an annual convention for prediction markets where many purists worry that Kalshi and Polymarket are undermining the technology's public good — At Mani

industrytechmeme

A Modular Vision-Language-Action Robotics Framework for Indoor Environments

arXiv:2606.31144v1 Announce Type: cross Abstract: This paper presents an integrated system for the CMU Vision-Language-Action (VLA) Challenge, designed to enable an autonomous agent to perform complex

agentsarxiv-cs-ai

A once in a lifetime crossover between @aiDotEngineer and @FIFAWorldCup is happening in SF! 100 speakers from AIE got the golden pass to see…

A once in a lifetime crossover between @aiDotEngineer and @FIFAWorldCup is happening in SF! 100 speakers from AIE got the golden pass to see team USA kick Bosnia's ass!! Thanks @swyx @liamcbride and T

toolsswyx--x

A Realistic Protocol for Evaluation of Weakly Supervised Object Localization

model-releasesarxiv-cs-cv

arXiv:2404.10034v3 Announce Type: replace Abstract: Weakly Supervised Object Localization (WSOL) allows training deep learning models for classification and localization (LOC) using only global class-