[LOCKSUIT]

OpenAI's new release is brain dropping! Below is 2 tests I ran in their playground, I show it evolving the context. In the first test, the input is written by me, and then I feed it in, then I feed in the output, repeat. In run two, same, each is separated by 2 newlines just to make it clear which is the next one I fed into the AI. My context used for all submits to the AI was "Improve the ideas written.". It surprised when when I realized it would not only edit the text, but also elongate the story. Here I thought it would need to do both still using some hack.

Here is my tests I ran: https://justpaste.it/4rir0

[LOCKSUIT]

1 more set of tests: https://justpaste.it/45ksq

[LOCKSUIT]

In comparison to my posted test of the new edit GPT-3 above, I tried older gpt3s.....hmm they can do the edit thing too, maybe only slightly worse due to not having double sided context, I didn't analyze my tests below much yet but looks like it:
https://justpaste.it/86nnv

[LOCKSUIT]

I tried more tests with davinci original GPT-3 just now, meh, it can be hard to impress I guess, it gave a great response (first attempt) to one of my tests look. My prompt ends at the 3rd quote (opening quote):

"I was walking into a store and bought an apple. I walked out onto the road and was hit by a car, my apple fell onto the road."

Here is the same story above but edited so that the person does not die:

"I was walking into a store and bought an apple. I walked out onto the road, I heard a car coming and quickly moved out of the way, but my apple fell onto the road."