121 sats \ 6 replies \ @kr 27 Apr 2023 \ on: SN feature bounty: chatgpt summaries of articles meta
this is a really interesting discussion - and one that all social platforms will probably encounter at some point.
i don’t have a strong opinion yet, but appreciate the back and forth perspectives so far.
on the one hand, i agree with @darthcoin that this has the potential to dumb down the site and reduce people’s ability to think critically… I think that’s one of SN’s superpowers right now. Lots of smart people willing to put in the work to get to the bottom of important issues.
on the other hand, i hear @k00b and also get frustrated with some of the low effort, inaccurate link titles that already fill up the recent tab.
what if this feature was off by default and could be turned on by users for a 100 or 1000 sat fee on each link?
what if this feature was off by default and could be turned on by users for a 100 or 1000 sat fee on each link?
This sounds really good! Was also thinking about "default off" but adding a fee sounds even better since the API isn't free.
However, do you mean the user who posts the link can turn it on or is it per user who views the link?
Making it a one-time fee for the user who posts the link makes sense from a "someone must pay"-perspective and since it's only one API request.
But do these users have the incentive to pay? Shouldn't the users who want to read the summaries pay? But multiple users would then pay even though it's just a single request for SN...
Maybe it could be a interactive prompt?
Then users can really leverage the power of AI: They can also start asking questions in natural language and get responses.
reply
good points. i haven’t yet thought through all the incentives, but figure there should be some money component since the API requests aren’t free.
the interactive prompt idea is a good one too, especially as SN’s content library grows larger. Imagine being able to ask the Reddit AI bot questions about any topic…
reply
A prompt on first iteration might be too complex though.
So I think just a well adjusted fee for every user who wants a summary could be good enough for now.
Would be really interesting to see how many users really would pay for this (depends on the fee of course which further depends on how much it costs SN).
reply
Then do rate limit the posts one can make. The more often a user creates new posts, the more sats per post they have to pay (it could increase exponentially). That could encourage people to focus on posting quality content rather than posting willy-nilly to try and see what sticks.
But no EvilGPT, please, for the love of God.
reply
Then do rate limit the posts one can make. The more often a user creates new posts, the more sats per post they have to pay (it could increase exponentially).
Afaik, this is already the case
reply
Thanks, I didn't know this was already the case.
But my stance on AI still stands: personally I'd rather avoid it as much as possible and prominently disclose when a particular piece of content has been generated by AI. And absolutely avoid the CrapGPT mainstream- and woke-biased AI. If a service I love is going to use AI regardless, at least use AI created and trained by bitcoiners and for bitcoiners, so that it generally shares my values (of course, that still doesn't guarantee a good outcome).
In this case, solving such a minor problem doesn't grant using AI, in my opinion.
reply