I *was* planning on doing a Twitch livestream of GPT-3 prompt testing today but GPT-3 still seems to be too hot of a topic nowadays and I need to wait for the hype to fade a little. That said, it now gives me the opportunity to build a better UI for GPT-3 livestreaming. 😃

A few issues with the OpenAI GPT-3 web UI: 1. It defaults to 0.9 temperature and 20 token generation which aren't good defaults. 2. You have to *manually* clear generated text to generate new text w/ same prompt. 3. No way to cancel a generation in progress. 4. NO DARK MODE.

Thanks for the feedback! On 1, What defaults would you prefer? Also on 2, have definitely thought about this — how would you prefer to continue an in-progress generation?

1. In my testing for creative output, 0.7 gives much better results than 0.9. 20 tokens isn't a lot either: 128/256 tokens would be better if API capacity is improved. 2. It's for the opposite use case; since it takes a few tries for good output, I have to clean back to prompt.