Search: “agents”

arXiv:2604.27372v1 Announce Type: cross Abstract: This paper investigates the continuous-time counterpart of the Q-function for entropy-regularized mean-field control (MFC) with controlled common nois

Cool paper from Meta FAIR. It's on self-improving LLMs but on the pretraining side. (bookmark it) Most LLM safety, factuality, and reasoning fixes get bolted on at post-training. By then, the patterns

arXiv:2604.27906v1 Announce Type: new Abstract: Persistent AI memory is often reduced to a retrieval problem: store prior interactions as text, embed them, and ask the model to recover relevant contex

arXiv:2604.16399v2 Announce Type: replace-cross Abstract: The widespread adoption of AI-assisted development tools in 2025 -- and the emergence of vibe coding, a practice of generating complete applic

arXiv:2604.27763v1 Announce Type: new Abstract: The emergence of Large Language Models (LLMs) offers a transformative interface for Web3, yet existing benchmarks fail to capture the complexity of tran

arXiv:2604.27865v1 Announce Type: new Abstract: Language models are saturating benchmarks for procedural tasks with narrow objectives. But they are increasingly being deployed in long-horizon, non-sta

arXiv:2604.27713v1 Announce Type: new Abstract: The risks posed by AI features are increasing as they are rapidly integrated into software applications. In response, regulations and standards for safe

arXiv:2604.27962v1 Announce Type: new Abstract: Designing mechanical linkages involves combinatorial topology selection and continuous parameter fitting. We show that language models can systematicall

On 5/5 @realDanFu and team will discuss DSV4’s hybrid attention and KV cache efficiency, should be a great session! Join us Tue 5/5: #DeepSeek-V4's hybrid attention + sparse MoE reduces KV cache up to

One week since the launch of GPT-5.5, and it’s already our strongest model launch yet. API revenue is growing more than 2x faster than any prior release, while Codex doubled revenue in under seven day

Our CEO @jerryjliu0 in @VentureBeat , on what's actually changing in the LLM stack: 'We've really identified that there's a core set of data that has been locked up in all these file format containers

arXiv:2604.26968v1 Announce Type: cross Abstract: Key-value (KV) cache memory management is the primary bottleneck limiting throughput and cost-efficiency in large-scale GPU inference serving. Current

The new Grok comes in below the latest Chinese open weights models, Grok 4 was at the frontier when released. (& Artificial Analysis: please stop using GDPval-AA which is not a useful test of anything

arXiv:2604.28076v1 Announce Type: cross Abstract: Large Language Models (LLMs) have advanced Table Question Answering, where most queries can be answered by extracting information or simple aggregatio

When training Grok 4.3, we spoke directly with devs and businesses to understand what they actually needed: a model that’s fast, affordable, and great at tool calling. The result is a daily driver tha

arXiv:2604.25963v1 Announce Type: new Abstract: Vehicle platooning has attracted increasing attention as a promising approach to improve traffic efficiency, energy consumption, and roadway safety thro

arXiv:2510.08049v3 Announce Type: replace-cross Abstract: Although Large Language Models (LLMs) exhibit advanced reasoning ability, conventional alignment remains largely dominated by outcome reward m

This post by Ethan Mollick likely discusses how advanced AI systems like hypothetical future GPT versions could autonomously execute complex, real-world tasks with minimal human direction, using a hum

arXiv:2604.26103v1 Announce Type: cross Abstract: All current LLM serving systems place the GPU at the center, from production-level attention-FFN disaggregation to NVIDIA's Rubin GPU-LPU heterogeneou

Enterprise software has no shortage of AI enthusiasm, but the gap between AI potential and production-ready results continues to frustrate organizations investing in enterprise process automation. As

At @sequoia’s AI Ascent last week, @gdb told me something that stuck: in late 2024, AI wrote ~20% of @OpenAI's code. That number is now 80%. We also got into why human attention, not compute, is the r

arXiv:2604.26382v1 Announce Type: cross Abstract: Most enterprise document AI today is a pipeline. Parse, index, retrieve, generate. Each of those stages has been studied to death on its own -- what's

arXiv:2506.20876v4 Announce Type: replace Abstract: Technological progress has led to concrete advancements in tasks that were regarded as challenging, such as automatic fact-checking. Interest in ado

arXiv:2604.26173v1 Announce Type: cross Abstract: An effective way to scale up test-time compute of large language models is to sample multiple responses and then select the best one, as in Grok Heavy

arXiv:2604.26180v1 Announce Type: cross Abstract: With recent semantic query processing engines, semantic aggregation has become a primitive operator, enabling the reduction of a relation into a natur

arXiv:2604.26065v1 Announce Type: new Abstract: Generative motion prediction must satisfy three simultaneous requirements for real-world autonomy: high accuracy, diverse multimodal futures, and strict

ml-intern is fully on mobile now you can launch 8 A100s from your phone. while on the couch. while commuting. wherever I just did this while biking. same sessions as your desktop too — start a run on

arXiv:2510.08547v2 Announce Type: replace-cross Abstract: Towards the aim of generalized robotic manipulation, spatial generalization is the most fundamental capability that requires the policy to wor

arXiv:2604.26067v1 Announce Type: new Abstract: We present RADIO-ViPE (Reduce All Domains Into One -- Video Pose Engine), an online semantic SLAM system that enables geometry-aware open-vocabulary gro

arXiv:2601.06160v2 Announce Type: replace Abstract: Large Language Models (LLMs) often suffer from ''Reasoning Collapse'' on challenging mathematical reasoning tasks, where stochastic sampling produce

arXiv:2509.11295v2 Announce Type: replace Abstract: Developing effective prompts demands significant cognitive investment to generate reliable, high-quality responses from Large Language Models (LLMs)

// When to Retrieve During Reasoning // Pay attention to this one, AI devs. (bookmark it) Most RAG systems retrieve once, before the model starts reasoning. Large reasoning models like o1 and R1 don't

A million-token context window is not a strategy. 🛑 Our Head of DevRel, @RoieSchwabco , explains why dumping data is killing your RAG performance: 📍 One needle in a haystack? Easy. 📍 Multiple needles?

arXiv:2604.25203v1 Announce Type: new Abstract: Deploying guardrails for custom policies remains challenging, as generic safety models fail to capture task-specific requirements, while prompting LLMs

arXiv:2604.25676v1 Announce Type: new Abstract: Multilingual retrieval-augmented generation (mRAG) is often implemented within a fixed retrieval space, typically via query or document translation or m

arXiv:2604.24978v1 Announce Type: new Abstract: Enterprise deep research often fails to produce decision-ready reports due to uneven information coverage, context explosion, and premature stopping. We

arXiv:2604.25136v1 Announce Type: new Abstract: We propose Frictive Policy Optimization (FPO), a framework for learning language model policies that regulate not only what to say, but when and how to

GitHub employees fixed a critical remote code execution vulnerability in less than six hours last month. Wiz Research used AI models to uncover a vulnerability in GitHub's internal git infrastructure

arXiv:2603.01070v2 Announce Type: replace Abstract: Solving complex geometric problems inherently requires interleaved reasoning: a tight alternation between constructing diagrams and performing logic

IMO DeepSeek v4 demonstrated utter confidence and competence by not benchmaxxing, not focusing on some BS final run cost, not even spending inference-optimal compute. just showed up, demonstrated SOTA

arXiv:2604.24921v1 Announce Type: cross Abstract: Vision-Language-Action (VLA) models are a promising paradigm for generalist robotic manipulation by grounding high-level semantic instructions into ex

arXiv:2604.25122v1 Announce Type: new Abstract: We present M^3-VQA, a novel knowledge-based Visual Question Answering (VQA) benchmark, to enhance the evaluation of multimodal large language models (ML

arXiv:2604.24954v1 Announce Type: cross Abstract: We introduce Nemotron 3 Nano Omni, the latest model in the Nemotron multimodal series and the first to natively support audio inputs alongside text, i

arXiv:2604.25102v1 Announce Type: new Abstract: Typographic prompt injection exploits vision language models' (VLMs) ability to read text rendered in images, posing a growing threat as VLMs power auto

Recreated the app with ml intern to compare. Went faster than in cowork, some things are better (ex direct integration in the desktop reachy mini app), some things are worse (had to debut the install)

arXiv:2604.25720v1 Announce Type: cross Abstract: Despite strong performance of deep learning models in retinal disease detection, most systems produce static predictions without clinical reasoning or