I studiously avoided looking at any submissions relating to ChatCGPT for several months, when they started flooding HN last year. I thought 'Boring chatbot. Who cares?'

Then, at Christmas family get together, I was chatting to someone I hadn't seen in a year, who's also interested in "techy stuff" and he asked if I'd used ChatGPT. I gave the same "...boring chatbot.." reply and he explained how it could do more than that and could even help write code for you.

Back in post-Chrimbo world, I was trying to work out a way to sort through my huge archive of photos and collate info about EXIF locations and also which photos didn't have EXIF data. I remembered the Christmas convo, fired up ChatGPT and asked it 'How would I write <description of problem> in Go?' and it spewed out some code.

I ran it and it worked [-ish] but also generated a few errors. I pasted each error into ChatGPT and said 'Fix this error' and it amended the code. Rinse and repeat a couple of times, as other errors cropped up and I ended up with some functioning code to do what I wanted to do.

I then asked ChatGPT to translate the same code into Crystal snd then into V and it did the same [although admittedly I could never get the errors out of the V code --too small an amount of training data, methinks!].

Purists may not be too enamoured of such AI-generated code. But, for someone like me who is an ocassional dabbler: ie. interested in tech and occasionally hacks something together to automate <something> on the comp, but is in no way a professional programmer, ChatGPT was a revelation. I could have achieved what I wanted myself. But it would probably have taken me a day or two and lots of internet searching. With ChatGPT's help, I was able to knock together the basic framework in seconds and then spend a few minutes refining it to handle various errors.

I was just talking to a VP of engineering about ChatGPT, the API and the like and she was thinking about cutting down the amount of engineers at her company.

Something like this came up in our discussion and it looks like it would make her decision easier unfortunately.

why text-davinci and not code-davinci?