Simon Willison

Found the prompts for the excellent new @IceCubesApp image description generation feature - it's using GPT-4 vision and sending the image with the prompt:

> What’s in this image? Be brief, it's for image alt description on a social network. Don't write in the first person.

It sets max_tokens to 50

https://github.com/Dimillian/IceCubesApp/blob/4f9e23296fa9c8abb812bc24f0f9a1ce0c86b28a/Packages/Network/Sources/Network/OpenAIClient.swift#L86-L101

IceCubesApp/Packages/Network/Sources/Network/OpenAIClient.swift at 4f9e23296fa9c8abb812bc24f0f9a1ce0c86b28a · Dimillian/IceCubesApp

A SwiftUI Mastodon client. Contribute to Dimillian/IceCubesApp development by creating an account on GitHub.

GitHub
Dec 6, 2023 at 7:35pmWeb
Show threadThomas RicouardDec 6, 2023
@simon @IceCubesApp indeed :)
Show threadDominiqueDec 6, 2023
@dimillian @simon @IceCubesApp will this work on a video anytime soon?
Show threadSimon WillisonDec 6, 2023
@apexnasi @dimillian @IceCubesApp GPT-4 doesn't have the ability to analyze video yet - you can send it a bunch of static frames from a video and ask it to stick something together from those but that's both expensive, slow and unlikely to produce great results
Show threadDominikDec 6, 2023
@simon Does max_token affect the answer being generated, i.e. lower max_tokens = shorter answers? Or is the response just cut off after reaching the limit?
Show threadSimon WillisonDec 6, 2023
@dom as far as I can tell GPT-4 does somehow take the max tokens into account when deciding what to generate
Show threadIan WoottenDec 6, 2023
@simon @IceCubesApp do you collate any of these found prompts anywhere on your blog?
Show threadSimon WillisonDec 6, 2023
@iwootten @IceCubesApp No, but I've been thinking it would be really interesting to build a collection of these somewhere - especially a collection that noticed when people changed their prompts