This is the best summary graphic for my training cycle that I've come up with so far.

data and a bit of exploration over here: https://observablehq.com/d/a2c0b7952cab9a02

I think I'm going to limit my posting on this account to strictly data visualization stuff, and move almost all of my posting to @llimllib , so follow me there if you want.

I'm not transferring there because I'm still going to keep this account alive.

I wish I had a billmill.org mastodon but I'm also not willing to do the work to set one up.

Even as a skeptic, this is neat.

I ask gpt to write me a webpage that displays a floating hexagonal prism, and it nails it. This is just a screenshot, the camera is rotating around the object.

(I get that this is a near-ideal circumstance for gpt, basically a slight modification to every first tutorial. Still, I think it's neat)

I cleaned up my command line interface for gpt

- added proper command line arguments and a usage message
- added a second command for image generation queries

Then made it into its own repository: https://github.com/llimllib/gpt-bash-cli

I'm still not at all a huge fan of these models, and of OpenAI the organization especially, but basically I feel like I need to be able to use them to evaluate them properly.

syntax highlighting theme designers consistently make the color of code comments too dim. My preference is to use color to distinguish code from comment, and increase the contrast to make it more readable.

For example, default github gist syntax colors comments in desaturated light grey, versus a darker green option:

I wrote a bash function to make API calls to openai so you can play with chatgpt in your command line. If that's your sort of thing, go ahead and steal any part you find useful:

https://gist.github.com/llimllib/b4e7428e434d7161b02d498421f651b9

included is an example session of asking gpt to write the same script