Notices by CubeRootOfTrue (cuberootoftrue@mathstodon.xyz)

@csgordon Let's say I write a paragraph of text, show it to ChatGPT, tell it to fix errors, and then publish the result (which generally has better grammar and flow, you also add that to the prompt). In my experience, LLMs mostly just repeat what I said, with longer sentences and different words, but it's still the idea I meant to write about. I have an automated editor. You don't normally give authorship to the copy editor, right? I'll admit it's possible to go too far---You still have to copy edit the result you get from the LLM, sometimes they are just wrong because you screwed up the context and it thought you were talking about something else. It's a tool. Tell it to up its Professionalism.

@julesh oh, marshmallows to you

@julesh binary logic proves too much
a -> (b -> a) is BS but binary logic thinks it's fine