Enjoying this paper about the language we use to describe what LLMs are doing (shared by @AnthonyBaker@mastodon.social) https://arxiv.org/pdf/2212.03551.pdf https://i.imgur.com/S9Fzoxh.png

Improving Shell packaging in the Nix ecosystem. Looking for the seams. I code/write/edit/read and give a fuck as a service. 

I like it because it's a good nudge to consider to what extent or how frequently the mode *we* are operating in is just text prediction. To this end I think it's telling how easy it is for ChatGPT to pass our smell-test before we really start breaking down what the language means.

The sense in which it is ~writing seems to be well-matched for the sense in which we are ~understanding before we engage critical mode.

I see this in writing/editing, too. Even good technical editors can let a lot of shit slip by until they hit something that smells bad enough to knock them into critical mode.