Ostap Hembara

@karpathy hmm, for the longer running experimentation loops RLMs might be very helpfull

I packaged up the "autoresearch" project into a new self-contained minimal repo if people would like to play over the weekend. It's basically nanochat LLM training core stripped down to a single-GPU, one file version of ~630 lines of code, then: - the human iterates on the prompt (.md) - the AI agent iterates on the training code (.py) The goal is to engineer your agents to make the fastest research progress indefinitely and without any of your own involvement. In the image, every dot is a complete LLM training run that lasts exactly 5 minutes. The agent works in an autonomous loop on a git feature branch and accumulates git commits to the training script as it finds better settings (of lower validation loss by the end) of the neural network architecture, the optimizer, all the hyperparameters, etc. You can imagine comparing the research progress of different prompts, different agents, etc. github.com/karpathy/autor… Part code, part sci-fi, and a pinch of psychosis :)

@Sentdex What was the correct behavior for the robot? Dodge the flying chair? Looks like office usecase

testing robot policies has never been so much fun

@oprydai Z going up is some dark shit tbh Y is up

if you are either of these people, I want to be friends with you.

@jaredsuniverse The best way to put it - progress is progressing, bought like million things and do some pre-coding and calibration, once everything gets delivered

@jaredsuniverse I am trying to build an s100 arm that will play some chess with me

@ThePrimeagen soon keyboards will sit under the glass in museum, cause people will just vibecode with voice

Kinesis Advantage 2 Where I started

@janusch_patas I feel like with 3D there so many things that can go wrong, different camera and points convention, different projections and spaces. Still need a lot of plotting to figure out what went wong

It’s over!! We are cooked.

@AntonioSitongLi put a pokemon stickers on, fooled me so many times

Does anyone know how to make a robot look less evil?

@akbuilds_ yeap, like robotic arms mounted to the table

@ostapagone you mean robotic arms? or meant literal hands?

Went from 480 → 3,100 followers in one month. Still processing it. If you're building in: AI No-code Vibe coding Indie hacking Say hi 👇 I’m following serious builders

@akbuilds_ I want to build these github.com/TheRobotStudio…

@ostapagone bunch of? Enlighten us.

@cviklihamar @X planning on building some robots

Dear @X algo, I am looking to connect with people interested in: - Builders - AI Tech - Full Stack - Hackers - Building in Public Say hi & let's grow together 👋

@ErikVoorhees @openclaw wrote a small npm package that installs openclaw in docker with file sharing and also ability to connect to the browsers inside the docker. Use it to run multiple claws on ur local pc Windows, MacOs or Linux: github.com/ostapagon/open…

@JulianGoldieSEO wrote a small npm package that installs openclaw in docker with file sharing and also ability to connect to the browsers inside the docker. Use it to run multiple claws on ur local pc Windows, MacOs or Linux: github.com/ostapagon/open…

OpenClaw just got 100x more powerful overnight. There’s a new “Awesome OpenClaw Skills” vault with 2,868 free skills. Here’s how it works. → Skills are like apps for your AI assistant → You install only what you need → Categories cover marketing, coding, automation, research → Skills add real capabilities like web search, booking, websites → You can install in 1 click from GitHub Save this video, you’ll turn OpenClaw into a real business system. Want the SOP? DM me. 💬

@BentoBoiNFT wrote a small npm package that installs openclaw in docker with file sharing and also ability to connect to the browsers inside the docker. Use it to run multiple claws on ur local pc Windows, MacOs or Linux: github.com/ostapagon/open…

It took me 1hr 43mins to set up my Mac Mini + OpenClaw as a complete noob Alex (my AI agent) is live and running, but I still have no idea what i’m doing really lol Time to find out if this thing is actually as powerful as everyone says 🫡

@rileybrown wrote a small npm package that installs openclaw in docker with file sharing and also ability to connect to the browsers inside the docker. Use it to run multiple claws on ur local pc Windows, MacOs or Linux: github.com/ostapagon/open…

I spent 200 hours testing OpenClaw, trying to find the perfect setup... My biggest takeaway: Keep your agents focused, and build a team. Here's a video. 00:00 Intro 00:48 Perplexity Computer and Manus 02:43 OpenClaw 03:52 Too many Skills to my first AI Agent 06:09 People want an employee with Intent 07:35 Testing Narrow AI Agents 08:33 Narrow Agent Example (YouTube Agent) 11:32 A Team of Narrow Agents... Why? 16:27 In Summary

@elie2222 what is the tps and what is ur setup?

HOLY MOLY I just tested Qwen 3.5 9b and it's the future 8GB of RAM. Less than Chrome

been playing @perplexity_ai Computer and its kinda wild - 19 models coordinated, subagents, cloud sandbox, persistent memory, runs for hours without babysitting asked it to build me a 3d nn architecture visualizer (think netron + bbycroft.net/llm). it parses pytorch text, onnx, safetensors, gguf files and renders them in three.js with collapsible repeated layers

@timourxyz even he is shaking a bit saying these news

My droid is monitoring the situation and giving me updates. Access to full dashboard with polymarket, latest headlines, asset prices, etc Improve your situation monitoring with open source robotics.