Nishan Mills

A great leader is motivated not by power but by compassion. Therefore he can do nothing but make himself a servant to those whom he rules. Such a leader is unequivocally respected, and loved for loving.” ― Richelle E. Goodrich #CoupLK

This is such a good post. orchidfiles.com/im-tired-of-ai…

The Head of Claude Code at Anthropic hasn't written code by hand in months. In 2 days he shipped 49 full features. 100% written by AI. He just dropped a 30-minute talk on exactly how he does it. More valuable than any $500 vibe coding course. Bookmark it.

Instead of watching Netflix, watch this 1-hour Yale lecture by Professor Ben Polak. It will change how you think about decisions in negotiations, business, and everyday life.

Introducing Project Glasswing: an urgent initiative to help secure the world’s most critical software. It’s powered by our newest frontier model, Claude Mythos Preview, which can find software vulnerabilities better than all but the most skilled humans. anthropic.com/glasswing

I made an interactive blog post about how JPEG image compression works: sophielwang.com/blog/jpeg

Important nuances to the story from Palli Thordarson Director of UNSW RnA institute x.com/pallithordarso…

Sorry to be the downer because this is an impressive story in some senses. But it is ~trivially easy to make a single mRNA vaccine. It's not hard. I cure mice of various cancers with various therapeutics all the time. I've made mice lose more weight in a month than tirzepatide does in a year. What is hard and expensive is proving its BOTH safe AND effective **in a randomized and controlled study in humans** while ALSO manufacturing it at clinical scale and grade. I am happy for this man and his dog. It is impressive. But y'all are overhyping it.

Proud with @UNSWRNA to have been involved & making the mRNA-LNP for Rosie. There are nuances here that the thread below misses but nevertheless, the intersection of RNA technology, genomic & AI poses an opportunity to change the way do medicine and make access more equitable 1/8

This is the gentleman who used AI tooling such as ChatGPT and Alphafold to help a team of scientists synthesis a custom cancer vaccine for his dog. #personalisedmedicine #AI

A lot of people have been asking if this can be done for their dogs and for people. I'm speaking with everyone involved to see what is possible here. If you would like to be involved, please complete the following Google form: bit.ly/4bkaowg

Cool use case of AlphaFold, this is just the beginning of digital biology!

I spoke about agentic engineering at the Pragmatic Summit last month, in a fireside chat hosted by Eric Lui - here's the half hour video plus highlight quotes and extra notes from our conversation simonwillison.net/2026/Mar/14/pr…

It took another two months but Chrome 146 is out since yesterday! And *that* means: with a single toggle, you can expose your current live browsing session via MCP and have your CLI agent do things in it. Aaand I have been waiting to deal with my LI connects until this moment.

thanks to @karpathy , now i have cracked the mystery why my agent doesn't follow my instruction closely enough.

a student took the ELO rating system from chess ran it through 95,491 tennis matches over 43 years, and trained an XGBoost model that predicts winners with 85% accuracy he tested it on the Australian Open 2025 completely outside the training data 99 out of 116 matches correct called every single Sinner win through the entire tournament the champion, before the first ball was hit no team, no funding, a laptop and free CSVs from the internet this is the best breakdown of a real sports prediction model I've seen study it or feed it to your AI agent

so you're telling me i can now... embed a video embed a voice memo embed a PDF embed an image embed text ...all in the same space? with one model? and search across all of them with a single query? time to rebuild everything.

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 :)

@karpathy Very inspiring as always! We are also open sourcing part of our infra on automated research for Gemini to evolve itself at github.com/google-deepmin… More complex than the nanochat setup but closer to SOTA LLM pre/post-training while staying as minimal as possible. More on the way.

🚨 BREAKING: Someone just rebuilt the entire AI assistant stack in Zig. It's called NullClaw. The binary is 678 KB. It uses ~1 MB of RAM. It boots in under 2 milliseconds. No runtime. No VM. No framework. No garbage collector. Just raw Zig. Here's why this is absurd: → OpenClaw needs a $599 Mac Mini and 1 GB+ RAM → NanoBot needs 100 MB+ RAM and Python → PicoClaw needs 10 MB RAM and Go NullClaw runs on a $5 board with 1 MB of RAM. Same functionality. 0.1% of the resources. Here's what's packed into that 678 KB: → 22+ AI providers (OpenAI, Anthropic, Ollama, DeepSeek, Groq, etc.) → 13 chat channels (Telegram, Discord, Slack, WhatsApp, iMessage, IRC) → 18+ built-in tools → Hybrid vector + keyword memory search → Multi-layer sandboxing (Landlock, Firejail, Docker) → Hardware peripheral support (Arduino, Raspberry Pi, STM32) → MCP, subagents, streaming, voice, the full stack Here's the wildest part: Every subsystem is a vtable interface. Swap any provider, channel, tool, memory backend, or runtime with a config change. Zero code changes. It even encrypts your API keys with ChaCha20-Poly1305 by default. 2,738 tests. ~45,000 lines of Zig. Zero dependencies beyond libc. 100% Open Source. MIT License.