Live data from Hacker News

GPT-3 Creative Fiction

gwern.net

101–103 of 103 posts

Re: GPT-3 Creative Fiction

#101
post #36

Holy crapoly. The quality (in both senses) of the output given an appropriately constructed prompt is incredible. I wonder if it's possible to get it to do the opposite of summarizing, ie. give it a plot summary and have it expand it into a fleshed out story that conforms to the summary...

I wonder if it's possible to get it to do the opposite of summarizing That's pretty much its main job: you provide a prompt and it writes a story about it.

A story prompt resulting in a story is looser than what I meant. Can it elaborate on a plot summary (that includes spoilers) and remain faithful to it? eg. given an outline of O'Henry's Gifts of the Magi, write an equivalent story that still concludes with the protagonists both having sold their prize possesion to buy a now useless gift for the other? Or will the plot get lost in the weeds?

Re: GPT-3 Creative Fiction

#102
post #39

Holy crapoly. The quality (in both senses) of the output given an appropriately constructed prompt is incredible. I wonder if it's possible to get it to do the opposite of summarizing, ie. give it a plot summary and have it expand it into a fleshed out story that conforms to the summary...

It doesn't seem to work quite as well. For single sentences, GPT-3 output usually hangs together pretty well. For longer stretches of text, often there are internal inconsistencies that are jarring when you read it, or parts that don't quite make sense when you read it as a whole.

Thanks for the reply. Aside from the coherence and consistency issues you've noted, does the output actually conform to the specified plot summary, or does it deviate from it?

Re: GPT-3 Creative Fiction

#103
post #16

I'm gonna put forward the very view that gwern repeatedly argues against: "but... it's not understanding ." So far I see no evidence that this thing or anything else like it has any actual understanding, any model of the world. Indeed it can't as it possesses no sensory apparatus. It's not embodied. It doesn't experience anything. I'm not sure the OpenAI folks would argue with me, but it seems Gwern asserts that this…

How could we accurately measure understanding? Honest question, because i am curious.
Post reply on HN