Live mentions
Real-time mentions from X plus the latest from Reddit and Hacker News. Updated every 10 minutes
Lightweight web GUI for managing vLLM model serving with speculative decoding
I saw nobody was doing it. I've never shipped anything like this but here you go: “SpecServe” a web GUI for @vllm_project serving with speculative decoding. One-click downloads, Docker mode, DFlash drafts. Open source tonight. https://t.co/kFA6B0JGaN
An in-the-wild benchmark for AI agents in the OpenClaw Environment.
🚀 Introducing WildClawBench: a benchmark for real-world, long-horizon agent evaluation. While the benchmark was first released in March 2026, we're excited to share the full paper today. 🔗 Paper: https://t.co/tdFR0bRNm6 🔗 Code & tasks: https://t.co/VW6uL2txtE
Read-only on-chain battle history viewer for Stomp.gg on MegaETH.
Built a tiny read-only @stompdotgg history viewer for onchain game on @megaeth Paste any address, see W/L, teams, move sequences. No wallet, no signing. Just public Engine events via Blockscout/RPC. Magic of onchain game! 🧙🧙♂️🧙♀️ https://t.co/HroT0xF73u Issues/PRs welcome!
Numerical Understanding in LLMs/MLLMs
Numerical understanding is an important capability for LLMs/MLLMs. I’ve started curating relevant papers here: Awesome Numerical Understanding Papers https://t.co/0PuxMdS6D1 Suggestions, missing papers, and PRs are welcome.
Lucebox optimization hub: hand-tuned LLM inference, built for specific consumer hardware.
Lucebox: hand-tuned LLM inference for consumer hardware — rewritten kernel by kernel. - Megakernel: Qwen 3.5-0.8B at 1.87 tok/J on a 2020 GPU, matching Apple M5 - DFlash 27B: Qwen 3.5-27B at 130 tok/s on a single RTX 3090 259 ⭐ https://t.co/6Rta9ArKzC
🚀 One-command requirement development flow for Claude Code - Complete workflow system with sub-agents, quality gates, and intelligent automation
做完计划都要 cc-review两轮，结果每次 Codex 都可以 Review 出计划的问题。怪不得 Token 消耗量大。 https://t.co/xfLjDWH4Xf
FreeCAD_FrameForge-Modified 是FrameForge的mod版本，主要是建立型材框架，对应原版新增了一些功能。并且对中文汉化
freecad Isolate https://t.co/nDUuxWdVgp https://t.co/aKIPi1ibDn
Let your agent talk to you. Through any speaker. In any room. 让你的 AI agent，在任何房间的音箱上，开口向你汇报。AirPlay 2 · DLNA · Local TTS. One command, every speaker. AirPlay 2 · DLNA · 本地 TTS。一条命令，所有音箱。https://t.co/rE9CZEaBIk
Replace Ads with They Live style slogans
They Live (1988) inspired Adblocker https://t.co/x9i61UxjYk (https://t.co/WbClA9mADS)
🪄 Create rich visualizations with AI
微软研究院免费神器 Data Formulator 上线，能替代每月70美元的Tableau席位！ 你只需用大白话描述想要的图表，AI就自动生成、自动清洗数据、自动写代码。支持截图识别、CSV、实时数据库、MySQL、PostgreSQL…30种图表类型，还能本地Ollama跑。 https://t.co/jEEfyGu3NW
A list of articles, videos, and tools related to the use of AI for OSINT.
A nice list of AI OSINT articles. Includes my GROK for OSINT article, as well as a few of the Anti AI takes that circulated a while back. https://t.co/MEWaSazSiH
We're releasing Sub-JEPA 🌐 LeWM makes end-to-end JEPA world models stable with Gaussian regularization. Sub-JEPA asks: what if the full-space Gaussian prior is too strong? Code: https://t.co/T30QZOlBlS #WorldModels #JEPA #RepresentationLearning #MachineLearning
Emotion Image Attachment Plugin for OpenClaw — Auto-classify emotions via LLM and attach matching images to Discord messages
Ever gotten mad every time your agent makes a mistake? It hits different when it has a face. Let your agent respond to you as a character. https://t.co/hsDe0CY4sq https://t.co/MTKBaX7HdW
multi platform circuit layout and schematic drawing tool
Circuit layout and schematic tool for DIY guitar wiring https://t.co/RBAtu8NSIO https://t.co/Y5RYs0C9Ws
A professional-grade, privacy-first image processing engine designed to convert SVG vectors into high-quality raster images (PNG, JPG, WebP) with zero server-side uploads.
DevDouble2/isvgtopng-engine: A professional-grade, privacy-first image processing engine designed to convert SVG vectors into high-quality raster images (PNG, JPG, WebP) with zero server-side uploads. https://t.co/Z9yRZPaTV7
Bullet Physics SDK: real-time collision detection and multi-physics simulation for VR, games, visual effects, robotics, machine learning etc.
🐾 Bullet Physics SDK: Real-time Collision Detection and Simulation Explore VR, games, robotics, and machine learning with realistic physics! Learn more: https://t.co/Uo4LMLjirk Follow @repocatai_git for more AI & robotics projects 🚀 https://t.co/8gL85Iea3e
Transparency documentation
New on https://t.co/qOlGy0daG3: Amazon Bedrock AgentCore Payments enables autonomous agents to make real API payments. Circle's Agent Stack brings USDC micropayments to agentic workflows. Plus: how to run Claude Code parallel sessions. https://t.co/oAovJElmFZ
PostBot 内容同步助手 一款开源的多平台内容同步分发生产力工具。 支持将文章、笔记、动态、图片、视频、音频等内容，一键同步发布至主流媒体平台。覆盖微信/微博/今日头条/小红书/知乎/百家号/企鹅号/视频号/抖音/快手/哔哩哔哩（B站）等国内主流媒体平台，可轻松扩展兼容 X（Twitter）、Facebook、Instagram、TikTok、YouTube、LinkedIn 等国际媒体平台。
最近发现一个挺实用的开源项目 PostBot，适合做多平台内容分发。 如果你平时需要把图文/视频同步发布到多个平台，这类工具能省不少重复操作时间。 项目地址我放评论区/下方了，有需要的可以自己看看。https://t.co/4s8hQ4ryXL #开源项目 #GitHub #效率工具 #自媒体运营 #AI工具
an evaluation protocol for simulated adversarial trajectories
Attending MLSys 2026 🦾! Come visit my poster for YPS where I’ll be presenting my work on SAT-Eval https://t.co/zCVfDPjcgd https://t.co/u0BQuxVHr0
We are seeking beta testers for a new offloading media plugin for #wordpress https://t.co/u5MRCl2bvF #beta #saas #wordpressplugin #offloading #media
The platform for shipping, governing AI agents — framework-agnostic, multi-runtime, GitOps-native, self-hosted.
@OpenAI just dropped a bombshell: a majority-owned Deployment Company backed by $4B+ and instantly acquiring Tomoro + 150 battle-tested engineers. Why go all-in on deployment consultants instead of just shipping more models? Because the real AI bottleneck isn’t building, it’s shipping.
🤖 A client-first, server-capable, fully type-safe router and full-stack framework for the web (React and more).
TanStack NPM Packages Compromised https://t.co/HgEIiu4lP5 (https://t.co/RFHH5rnGdE)
岐黄脉镜 - AI中医脉诊分析系统 | TCM Pulse Diagnosis with Apple Health Data
AI把脉把握诊断成了48岁😭还让我少喝咖啡！ 28种脉象从你的Apple Watch的ECG数据里分析得到，还可以给出调理建议，我感觉我确实需要少喝点咖啡了。GitHub在线体验&Skill：https://t.co/3qZYAtxLkp https://t.co/4oWIOJXx5l
Not every hackathon ends with a win. Some end with something better. Built ReelStudio at the @aitxcommunity x @OpenAI Codex Hackathon in Austin. Upload photos. Get a cinematic, beat-synced video in under 5 min. No videographer. No editor. Github: https://t.co/RRyTs8Fr6o https://t.co/AsKx1rpenC
A single CLAUDE.md file to improve Claude Code behavior, derived from Andrej Karpathy's observations on LLM coding pitfalls.
@Mnilax Triad review of rules one-line per reviewer: Karpathy reviewer: "A harness that omits evals is not a harness; it is a vibe." Consistency reviewer: "The doc is self-undermining — six rules contradict at least one other rule." Practitioner: "60% theater, 40% useful — and the theater rules will poison the useful ones.