With no ML background + just LLM help, Brad Masters trained a colour pop model from scratch The resultant model is 4,000 times (!) smaller than Flux Kontext + LoRA is to get similar results on such a task.

Sometimes I think about the Alcubierre drive; A theoretical concept that exploits general relativity (Einstein’s field eq.) to manipulate space instead of propulsion for faster than light travel To simplify: The ship is stationary, in a vacuum that contracts space in-front and

this is your friendly reminder that you can run Qwen 4b on 25 000 prompts at 16000 tok/s inputs in less then 4 minutes on a 0.80 $/hr rented Lambda cloud instance with NVIDIA A6000. That is 375 000 short-ish prompts / hr (for 0.80$).

im a genius i built a web scraping/automation tool that does everything in < 1.0seconds its 920% faster than browser base proof: benchmarked against browser base's own demo (watch the video) you can use this with literally everything. im a genius live right now @ parse .bot

real-time evolution simulator with digital organisms

LLM runs inside a PDF file

I don’t get it. If all of you can access infinite memory (only works with ChatGPT), and the only thing stopping it is the world not seeing the article below then copying and pasting it into their ChatGPT… does that mean OpenAI always had the ability to give us infinite memory

ᴛʜɪs ɢᴀᴍᴇ ᴡᴇ ᴘʟᴀʏ ✧ 🔊🎶

Self-Questioning Language Models: LLMs that learn to generate their own questions and answers via asymmetric self-play RL. There is no external training data – the only input is a single prompt specifying the topic.

Here’s a quick update on Sentradel, a new company I’m starting with Cameron Rowe #robotics #drones

The Chrome Altars of Extinction

analogmirage

And suddenly we’re so over

🚀I built Voltra, an npm package that lets you: ship custom Live Activities, Widgets, Dynamic Island + more with React. No Swift, Xcode, or extra JS runtime needed. ⚡ Hot reload ⛽ Server push ✻ AI coding agent-friendly 𝝠 @Expo + React Native ready Demo + must-read

Holy shit... Tencent researchers just killed fine-tuning AND reinforcement learning in one shot 😳 They call it Training-Free GRPO (Group Relative Policy Optimization). Instead of updating weights, the model literally learns from 'its own experiences' like an evolving memory

Decide whether you want to be an entrepreneur or cosplay as one. Cosplaying is posting about your lifestyle, chatting with other founders—networking, telling stories, being active on social media, praising and pleasing your heroes. Being an entrepreneur is doing all the hard

🚀 Introducing rLLM v0.2 - train arbitrary agentic programs with RL, with minimal code changes. Most RL training systems adopt the agent-environment abstraction. But what about complex workflows? Think solver-critique pairs collaborating, or planner agents orchestrating multiple