Robert Keus 👨🏼‍💻

@birk NYC

I'm seriously considering moving to 🇺🇸 NYC? In a heartbeat, but not ideal for work. SF? Great for work, but *feels* less great for family. Which areas would be?

@Zephyr_hg Nice! Can you try how this works with our GreenPT.ai api? I can give you free credits.

I never run out of content to post anymore. Built an automation that monitors 50+ news sources, scores articles for relevance, and writes social posts automatically. It finds trending topics in my niche before they explode everywhere else. Saves me 15-20 hours monthly and keeps me ahead of every trend. Comment "NEWS" and I'll DM it to you (must be following)

this was soooo much fun to organize and watch! we'll do another worldwide one in 2026 - raise your hand here to participate: kidsaicoding.com/kids-hackathon…

What happens when you put 1500 kids, the best Vibe coding tool, a great agency, two AI entrepreneurs, and the best organizer in one room? This video tells more than 1000 words. Lear more at KidsAiCoding.com. Thanks @lovable @brthrsagency , @Thom_Wolf, @antonosika and Fabian Hedin for making this happen ❤️

GitHub repo: github.com/karpathy/nanoc… A lot more detailed and technical walkthrough: github.com/karpathy/nanoc… Example conversation with the $100, 4-hour nanochat in the WebUI. It's... entertaining :) Larger models (e.g. a 12-hour depth 26 or a 24-hour depth 30) quickly get more coherent.

Excited to release new repo: nanochat! (it's among the most unhinged I've written). Unlike my earlier similar repo nanoGPT which only covered pretraining, nanochat is a minimal, from scratch, full-stack training/inference pipeline of a simple ChatGPT clone in a single, dependency-minimal codebase. You boot up a cloud GPU box, run a single script and in as little as 4 hours later you can talk to your own LLM in a ChatGPT-like web UI. It weighs ~8,000 lines of imo quite clean code to: - Train the tokenizer using a new Rust implementation - Pretrain a Transformer LLM on FineWeb, evaluate CORE score across a number of metrics - Midtrain on user-assistant conversations from SmolTalk, multiple choice questions, tool use. - SFT, evaluate the chat model on world knowledge multiple choice (ARC-E/C, MMLU), math (GSM8K), code (HumanEval) - RL the model optionally on GSM8K with "GRPO" - Efficient inference the model in an Engine with KV cache, simple prefill/decode, tool use (Python interpreter in a lightweight sandbox), talk to it over CLI or ChatGPT-like WebUI. - Write a single markdown report card, summarizing and gamifying the whole thing. Even for as low as ~$100 in cost (~4 hours on an 8XH100 node), you can train a little ChatGPT clone that you can kind of talk to, and which can write stories/poems, answer simple questions. About ~12 hours surpasses GPT-2 CORE metric. As you further scale up towards ~$1000 (~41.6 hours of training), it quickly becomes a lot more coherent and can solve simple math/code problems and take multiple choice tests. E.g. a depth 30 model trained for 24 hours (this is about equal to FLOPs of GPT-3 Small 125M and 1/1000th of GPT-3) gets into 40s on MMLU and 70s on ARC-Easy, 20s on GSM8K, etc. My goal is to get the full "strong baseline" stack into one cohesive, minimal, readable, hackable, maximally forkable repo. nanochat will be the capstone project of LLM101n (which is still being developed). I think it also has potential to grow into a research harness, or a benchmark, similar to nanoGPT before it. It is by no means finished, tuned or optimized (actually I think there's likely quite a bit of low-hanging fruit), but I think it's at a place where the overall skeleton is ok enough that it can go up on GitHub where all the parts of it can be improved. Link to repo and a detailed walkthrough of the nanochat speedrun is in the reply.

Kids vibe coding games 👾 with @Rosebud_AI in Singapore as a part of the global Kids Hackathon!

Yesterday was a blast. 25 kids were vibe coding for 3.5 hours, laughing, building, and especially sharing a lot of positive energy. They were unstoppable! I especially like this creation: 🍔 Burgers Invaders 👾 The concept is simple: shoot with ketchup at the burgers! cheeseburger-invade-galaxy.lovable.app Today day two! Kids AI Coding. 80 locations, 1500+ kids and 300 mentors 💪 And a special thanks to @lovable @Rosebud_AI

1500+ kids from all over the world are vibe coding games! Thanks @AnthropicAI for helping us bring game creation to kids across 6 continents! @Thom_Wolf, @robertkeus, @marine_caous

@rileybrown_ai Today we organize the World Wide Kids AI Coding Hackathon!

Lovable’s pivot to being a tool for children is actually smart.

Today is Kids AI Coding Day! Kids (1500+ 🤯) from all over the world, India, Singapore, Australia, the USA, Belarus, Canada, Belgium, France, Germany, Ghana, Egypt, etc. are going to vibe code today and tomorrow. Seeing kids have so much fun while exploring their creativity and learning how to use AI in a fun way is one of the coolest things to do. And, of course, a shout‑out to our partners @lovable and @Rosebud_AI.

In a few days, several thousand (!) kids around the world will join the first Worldwide Kids Vibe-Coding Hackathon! I’m so excited to see what they’ll come up with. During our first small event, I was mind-blown by how quickly kids picked up these tools and how they used them with the kind of unbridled creativity we can only dream of as grown-ups. You can still join! Contact @marine_caous she’ll get you set up in no time.

@zeeg So true

dont condone this kind of thing go to school, go to college, learn some things, experience some things, then decide what you want to do even if you're brilliant (you're probably not), you're making decisions that impact the rest of your life

When my daughter asked me how she could start coding with AI, I realized how important it is to give kids the right space to explore and create. That’s why we started the Worldwide Kids AI Coding Hackathon. In just 7 days, thousands of kids will be coding together in a global hackathon. In this video, @Thom_Wolf and I explain why we started this. If you want to be part of this journey, Marine Caous can get you set up quickly or checkout our website: kidsaicoding.com The versions in the Netherlands (Utrecht) at Brthrs Agency are already fully booked. P.S. It’s so hard to see yourself in a self‑made video. How are all these TikTok kids doing this? 😂

@sama What do you buy?

I mostly buy stuff from ChatGPT now, so I am excited for this new feature!

@wycats Terminal interface

When using Claude Code, what is your number one frustration? Something that comes up over and over again and, if fixed, would drastically improve your experience?

yet another amazing call with hosts for the kids-ai-coding hackathon on 10-11 October and we've just added our first African host. we now have hosts on the 5 continents you can still join at kidsaicoding.com/kids-hackathon…

@AlexandrePesant Okayish. Bugfixing good, UX bad.

GPT-5 in Lovable, how is it? Super interested in your feedback!

Host applications for the Kids AI Vibe Coding event this October are exploding 🚀 We’re already on 4 continents🌍 with amazing partners like LinkedIn, Bayer, Stanford, Cognizant & more Wanna bring this free 9–12 yo AI coding event to your city? Join → kidsaicoding.com/kids-hackathon…