Ai

From Andrej Karpathy:

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/autorese…
  Part code, part sci-fi, and a pinch of psychosis :)

Been trying to think of ways to leverage this. There have been some amazing examples so far:

• Tobi, founder of Shopify made their template system 53% faster (keep in mind this has been around 20 years)

• Cheng Lou, made a leap forward in UI engineering with a text measurement algo in pure TS called pretext (I can’t tell if he used autoresearch specifically, but the footprints are there in the repo)

What else is autoresearch going to push forward?

But when I started building with agents, it would break me out of what I was doing to go leave breadcrumbs in Roam so I could track what I had been up to. So I stopped. And honestly, it was something I noticed when I worked on meaty problems when programming too, I would get caught up in what I did and not take notes. To my detriment later.

I’ve also noticed this. Trying to get into flow with agents is harder.

Skill issue for me? Probably. But this is all so new, and changing so fast.

I like Björn’s idea here a lot. Definitely need to explore Beads!

If you asked me 5 months ago what software engineering would look like, I would not have been close to how it actually is today. Not even.

This file is my way of pointing my agents to my “taste” in terms of how I like to engineer software.

Things are changing so quickly, how we direct agents will too.

I like Scott's take here. It's not dead. Never was. Just changed.

Welp, that was fast.
From Sam Altman:

Peter Steinberger is joining OpenAI to drive the next generation of personal agents. He is a genius with a lot of amazing ideas about the future of very smart agents interacting with each other to do very useful things for people. We expect this will quickly become core to our product offerings. OpenClaw will live in a foundation as an open source project that OpenAI will continue to support. The future is going to be extremely multi-agent and it's important to us to support open source as part of that.

This has been the new hotness on the socials. But it's kinda just really good packaging for a local proxy to foundation models. Not saying it's nothing. Still might try it out. But I'm busy, yo.

Dario's (founder of Anthropic) essays are required reading.

oooo, boy going through this right now.

As a newcomer to a long-existing codebase recently, AI provides a lot of productivity gains.
You can choose how you use AI.
Choose wisely.

I like this definition better. As we learn more about how to manage our words passed to LLMs, I think the nouns we use should change along with it.

My most-used, most-recommended app, just went mobile with an iOS app.

It started as a shortcut for everything, but now with all the LLM models, is my go-to for quickly asking AI questions throughout my day.

I think Addy has the right take here.

You can avoid it and fall behind.

But if you go too deep down the rabbit-hole of letting AI write your code, you're asking for problems. Most importantly: skill atrophy.