My talk is on YouTube! youtu.be/sDEohTrNVvc?...

Exactly!

I have a small desktop heater that blows across my keyboard. It was only ~$20 and made a huge difference.

Discovered this because I was trying to look at the teleport docs and got a 500... figured they had to be up to something with Wire:Live happening haha

A couple big things I still want to add:

- Caching of generated text to avoid LLM calls for every reseed
- A better way to set config on the industry instance.

Hoping to knock out caching this weekend

You can submit feedback on Google results so I went ahead and did that. Hopefully it makes a difference.

Thanks! Glad to know I'm not the only one. Hoping to publish an early version of this in the next couple weeks

This is really neat

I don't! But maybe I should since I add a suffix to controllers and most other things

Also, I've changed the "string" method to "describe". Non-string data shouldn't be generated with the LLM so I don't plan to support anything else

On my M3 Pro MacBook, I'm getting fairly fast, and very consistent structured output with Ollama and the Llama 3.2 model

Something like this is now possible thanks to smaller LLMs and wider support for structured output. And, of course, all the work by @tjmiller.bsky.social and the community on Prism!