I've shared this link before, showing Riley Goodside's artful prompt-crafting, but it's again relevant to this topic: twitter.com/goodside/status/15818055038…... GPT-3 is perfectly capable of recognizing what kinds of things it will be bad at, and can be encouraged to generate machine-executable queries to fill in that gap. (note this is based on prompting GPT-3, not chatGPT, but the principles about what this language model is capable of apply) It can even go one level deeper - there's an example there of it generating a python script that uses the 'wikipedia' library to look up the date of death of the Queen, as a way to fill in knowledge it doesn't have. Tell it it can use the wolframalpha module to answer questions that involve complex units, quantities, or advanced mathematics, and it'll almost certainly do that too. One of the things I love is this reply to that tweet - twitter.com/JulienMouchnino/status/1582…... "How is it possible that GPT-3 understands what a human can compute in his/her head?" Riley's quick demo prompt that shows how well ChatGPT can guess whether particular mathematical results are easy to guess matches human intuition surprisingly well.

I recorded a demo of this same premise here: twitter.com/goodside/status/15626130289… Text completions of exotic forms of session/action logs are a seriously under-explored area. Here’s what happens if, instead of a text game, you do text completion on an IPython REPL: twitter.com/goodside/status/15818055038…

I can't speak for others, but I've personally been quite impressed by the dalle output. It creates things that would take me hours (if not days) to create, which no other tech I've tried has been able to generate. It feels like it can absolutely replace at least the stock photo industry. It's also terrific for things like blog photos if you don't have the time or talent to create something yourself, but want some creative control. Expansions like dream booth, which let you fine tune the system with your own submitted images are also quite amazing. Being able to give it just a few photos, and say things like "show me surfing in the ocean" and get a reasonable image back. _Much_ more broadly, this space in AI/ML with GPT3/Dalle is exciting because it feels kind of like what the internet was made for. There's too much data on the internet for any one person to ever meaningfully process. But a machine can. And you can ask that machine questions. And instead of getting just a list of references back, you get an "answer". Image generation is the "image answer" part of this system. It's an exciting space because it feels like these systems will affect large chunks of how we use computers. Here's a cool GPT3 "programming" example: twitter.com/goodside/status/15818055038… And here are some of my dalle uses I've been impressed by, that I feel is publish-ready: - labs.openai.com/s/nkOTLRWzjgQTe4QsgoWCh… - labs.openai.com/s/uSP55qRf1SqCbYTa2UDXX… - labs.openai.com/s/kO2purvEodK5UUxPIpL78… - labs.openai.com/s/2P1Mb75JbS1mmpyi86xwy…

Solving math problems and writing code are pretty similar. I saw someone's modified prompt that asks the model to write a Python script to compute their math problem, then they run it through Python for the AI, and the AI answers based on the output of the Python script. twitter.com/goodside/status/15818055038…

There's this GPT-3 prompt that does exactly this. It's not very good as of now, but looks super promising to me. [1]: beta.openai.com/playground/p/1PBppouAUt…... [2]: twitter.com/goodside/status/15818055038…

People love pointing out how bad GPT3 is at math - it’s just not remotely surprising, though. We all learned a long time ago that just patternmatching isn’t enough to solve math questions and we need to actually activate our brains and run algorithms to get valid answers. Even mathy people fall into this - we all know that some numbers ‘look prime’. And up to 100 that instinct can work really well (except for 91). But we all apply a heuristic for when to stop guessing and start mathing. But guess what! GPT3 can do that too! twitter.com/goodside/status/15684481284… This thread predates chatgpt - but it really shows that GPT can pretty reliably spot when winging it won’t work. And in this case, it could effectively be given the ability to whip out a calculator to figure it out. (Edited to add) Found the follow up thread where Riley went into more detail and enhanced the results: twitter.com/goodside/status/15818055038…...

Yes: twitter.com/goodside/status/15818055038…

This user crafted a prompt where ChatGPT is instructed to generate IPython to help it answer questions, which the user then runs for it, and gives the output back to ChatGPT, which then interprets it to create the answer to the question: twitter.com/goodside/status/15818055038… In other words it seems straightforward to connect ChatGPT to an external API.