All entries

We’re introducing the Cursor SDK so you can build agents with the same runtime, harness, and models that power Cursor. Run agents from CI/CD pipelines, create automations for end-to-end workflows, or

We're open-sourcing Hy-MT1.5-1.8B-1.25bit — a 440MB translation model that runs fully offline on your phone, supports 33 languages, and outperforms Google Translate. At 1.8B parameters, it matches com

We've open-sourced a few starter projects for you to build on: a coding agent CLI, a prototyping tool, and an agent-powered kanban board. Use Cursor to customize them for your use case: http://github.

Production low-latency autocomplete implementations employ diverse strategies including inference server optimization (tools like vLLM, llama.cpp, NVIDIA Triton), deployment choices (cloud APIs, on-pr

arXiv:2604.25132v1 Announce Type: new Abstract: Instruction-tuning datasets often contain substantial redundancy and low-quality samples, necessitating effective data selection methods. We propose an

AI-driven workflows are quietly redefining how work actually gets done. As organizations move beyond isolated automation, the emphasis is shifting toward embedding intelligence directly into the flow

Agentic enterprise is moving artificial intelligence from isolated tools into systems that actively run business operations. The transition reflects a broader shift in enterprise strategy, where organ

Dell Technologies’ AI Factory is emerging as a core building block for enterprise AI deployment, delivering full-stack infrastructure for training and inference since its 2024 launch. Yet as AI moves

arXiv:2604.25872v1 Announce Type: new Abstract: Training language models via reinforcement learning often relies on imperfect proxy rewards, since ground truth rewards that precisely define the intend

arXiv:2604.25213v1 Announce Type: new Abstract: OpenAI's GPT-Image-2 has effectively erased the visual boundary between authentic and AI-edited document images: a single number on a receipt can be rep

arXiv:2510.07499v2 Announce Type: replace Abstract: Recent Long-Context Language Models (LCLMs) can process hundreds of thousands of tokens in a single prompt, enabling new opportunities for knowledge

Replit's Agent feature can diagnose app issues by analyzing user-reported problems pasted directly into the platform, automatically identifying root causes and suggesting fixes. This tool streamlines

when you take an agent to production, you almost certainly need guardrails! deepagents makes it easy to add guardrails via middleware. guardrails can be used for 1. control flow, like model/tool retri

arXiv:2604.25611v1 Announce Type: new Abstract: Real-time automatic speech recognition (ASR) systems face a fundamental trade-off between transcription accuracy and computational efficiency, particula

On April 22, 2026, attackers hijacked Checkmarx's KICS security scanner and Bitwarden's CLI within hours of each other, both using the same command-and-control domain in a coordinated supply chain cam

arXiv:2603.14248v2 Announce Type: replace-cross Abstract: Large language model (LLM) web agents are increasingly used for web navigation but remain far from human reliability on realistic, long-horizo

arXiv:2604.25011v1 Announce Type: new Abstract: Reinforcement learning (RL)-based post-training often improves the reasoning performance of large language models (LLMs) beyond the training domain, whi

arXiv:2604.25012v1 Announce Type: new Abstract: Automated agentic workflow design currently relies on per-task iterative search, which is computationally prohibitive and fails to reuse structural know

arXiv:2604.25384v1 Announce Type: new Abstract: This paper presents a methodology for transforming raw Wikimedia dumps into quality textual corpora for seven South Slavic languages. The work is divide

A new report out today from Google LLC-owned cloud security company Wiz Inc. finds that artificial intelligence has shifted from experimental tooling to default cloud infrastructure, with 81% of obser

This post documents a Wordle game result where the player solved puzzle 1,774 in four attempts, with the final answer being a five-letter word with the pattern matching the green squares shown in the

Xiaomi MiMo-V2.5-Pro achieves multiple breakthroughs in the latest Arena rankings (Apr 26, 2026) 🔥 🏆 Text Arena (Expert) — #6 globally | #1 open-source model Also #1 among Chinese models, with Xiaomi

Yes, just having students “use AI to study” hurts learning (a helpful assistant is not a tutor), but using AI prompted to act like a tutor, especially with teacher support, seems to have large positiv

Yesterday, @xenovacom assembled Reachy Mini and this morning I created my first app for this new version in under an hour with Claude! Try it on your Reachy mini (or in simulation): https://huggingfac

You can now control both local and cloud ComfyUI with Hermes Agent with ease with the new built in skill `hermes update` and run /comfyui to get started ComfyUI is the most flexible, composable, and p

Nous Research has introduced a feature enabling Hermes agents to dynamically create and manage their own live to-do lists during task execution. This capability allows agents to better organize and tr

Jay Peters / The Verge: YouTube rolls out Full Multiview Customization to YouTube TV, letting users pin up to four live streams in a window, after debuting a limited version in 2023 — Custom multiview

arXiv:2604.25076v1 Announce Type: new Abstract: Many Multi-Agent Reinforcement Learning (MARL) agents fail to adapt properly to cooperating with agents trained with the same objectives but different s

Gary Marcus critiques Mark Zuckerberg's claimed focus on biology research, suggesting the financial commitment is minimal compared to his AI investments and stating that Zuckerberg's actual priorities

2026: the year of explosive AI innovation. @FireworksAI_HQ Co-founder & CEO Lin Qiao speaks on how the massive expansion of customized inference and models is creating speed of light production to sca

The 2026 Vercel AI Accelerator was a program designed to support and mentor early-stage AI startups, providing resources, mentorship, and potentially funding to help them grow and scale their products

NVIDIA and SLB are collaborating to build generative AI models using NVIDIA NeMo and NIM for subsurface exploration, production operations, and data management in the energy industry. This approach ac

arXiv:2604.22830v1 Announce Type: new Abstract: Pre-training is a general method that is used in a range of deep learning tasks. By first training a model on one task, and then further training on the

arXiv:2604.23935v1 Announce Type: new Abstract: Audio-based video object segmentation aims to locate and segment objects in videos conditioned on audio cues, requiring precise understanding of both ap

300,000 AI builders have already added their hardware to HF to instantly see what model they can run locally. To do so, go to https://huggingface.co/settings/local-apps and add your hardware specs. Yo

At Google Cloud Next ‘26, we announced that more than 50 Google-managed Model Context Protocol (MCP) servers are generally available or in preview, with more on the way. Why it matters: To move beyond

arXiv:2604.24149v1 Announce Type: new Abstract: Remote sensing images are frequently degraded by adverse weather conditions, particularly clouds and haze, which severely impair downstream applications

arXiv:2604.23458v1 Announce Type: new Abstract: The growing availability of online support groups has opened up new windows to study mental health through natural language processing (NLP). However, i

arXiv:2601.13288v2 Announce Type: replace Abstract: Production LLM systems often rely on separate models for safety and other classification-heavy steps, increasing latency, VRAM footprint, and operat

Thariq announced progress on optimizing the no flicker renderer with the goal of making it the default rendering option in an upcoming release. The post indicates that further updates regarding this d

A big problem with all AI at work punditry right now is that it all rests on data from the pre-agentic era (which is basically just now ending) and we have very little information about what has been

GM's Super Cruise hands-free driving system has reached 1 billion cumulative miles driven across nearly 750,000 vehicles , marking a major achievement since the feature's 2017 debut. Nearly half of th

arXiv:2603.07475v2 Announce Type: replace Abstract: Autoregressive (AR) language models build representations incrementally via left-to-right prediction, while diffusion language models (dLLMs) are tr

arXiv:2604.23049v1 Announce Type: new Abstract: AI agents are increasingly deployed to execute tasks and make decisions within agentic workflows, introducing new requirements for safe and controlled a

arXiv:2604.23045v1 Announce Type: new Abstract: Systematic biases in Global Circulation Model (GCM) outputs limit their direct applicability in regional planning, necessitating bias correction. Correc

arXiv:2604.22858v1 Announce Type: new Abstract: Liver cancer, especially hepatocellular carcinoma (HCC), imposes a substantial global disease burden. Accurate diagnosis and prognostic assessment direc