2 Comments

I've recently been experimenting with different AI tools like ChatGPT and Whisper, but it seems like Bard, the AI chat model from Google, might not be as accurate as they claim. I've turned to Antoine-hosted Claude from Anthropic instead, but requesting an API key feels like going through so much paperwork. Hopefully, these AI models will get better at math soon or misunderstand me the way their programmers misunderstand sarcasm. Oh, the joys of technological progress. Keep developing, folks.

Expand full comment
author

I think, like most models, they're good for what they were designed to do and increasingly bad the further you go outside that area. I think Claude and the others handle text much better than math because so much of predictive text is pattern matching anyway, so it will be great a repetitive jobs that involve a lot of boilerplate code. Whether the semi-mythical Q* algorithm will change their ability handle math we'll just have to wait and see.

Btw, the form to get an API key at Claude looks like a lot, but I didn't supply much info at all and it worked, maybe because I'm small enough to be considered harmless. :)

Your bigger point, though, about not giving up your day job just yet, I think is quite true.

Thanks for commenting!

Expand full comment