You are viewing a single comment's thread from:

RE: Steem as an AI incubator - again: The StoryChain [burnsteem100]

I think there would be major diminishing returns from trying to write and rewrite the same story, maybe it would work if the reading audience was huge and independent samples encountered each new instance but that's not really a good description of the Steem ecosystem. The idea would probably work better with fresh stories each day, maybe trying to learn from the set of recent responses. (Also, I don't think there's much audience for fiction here, and it seems like many people who do like fiction have issues with non-human-generated fiction so that would likely be an issue to navigate). Another variation to consider is to run it like a contest, and have N AI-authors that are independently self-evolving trying to compete with the N-1 others, then each one can use the rank-order of yesterday's stories as feedback for how to improve. I guess there would be context-frame issues, though.

Sort:  

I think there would be major diminishing returns from trying to write and rewrite the same story

I agree. I didn't make it clear, but I wasn't envisioning that it would be the same story rewritten every time, but just that there would just be some connection(s) between consecutive stories. Maybe characters would be repeated one day, and the setting would be repeated on another day, and so on. Especially with the LLM rewriting it's own prompt, there could be a lot of latitude for separate but connected story lines.

Another variation to consider is to run it like a contest, and have N AI-authors that are independently self-evolving trying to compete with the N-1 others, then each one can use the rank-order of yesterday's stories as feedback for how to improve. I guess there would be context-frame issues, though.

That might even emerge naturally, if multiple AIs are competing for rewards. The trick is to avoid drowning out the human presence, though. We need the scale to be big enough, but not too big. ;-)