Re:Philosophers on GPT-3
by hirundo on Thursday July 30, 2020 @11:07PM
(#24005259)
For instance a gpt-3 bot could make a high scoring HN commenter. Surely someone has experimented with that. Any preliminary results?
[ Parent
| on HN ]
Re:Philosophers on GPT-3
by renewiltord on Friday July 31, 2020 @02:35AM
(#24006251)
Tried with GPT-2. Got downvoted and never exited auto-flag.
The structure of GPT-3 makes this a bit hard right now - most people don't have access and even then it is limited, making feeding the articles and other comments in is difficult. This isn't a fundamental difficulty though, so I expect that we'll see it in the near future.
Re:Philosophers on GPT-3
by theontheone on Thursday July 30, 2020 @11:46PM
(#24005456)
I have access — it's actually quite easy to use? People have also built tools say, in python, that allow you to quickly test prompting (it's somewhere on tech/vc twitter...) Like you said there isn't fundamental difficulty, there are just far more interesting things to do
Re:Philosophers on GPT-3
by tangjurine on Friday July 31, 2020 @12:18AM
(#24005598)
>it's somewhere on tech/vc twitter...
?
> >it's somewhere on tech/vc twitter... ?
I'll translate: "It was posted on Twitter by someone who typically posts on tech and VC topics, and/or who typically interacts on Twitter with other accounts active WRT those topics. "
Alternatively: " The link was recently retweeted by various accounts that are active participants in tech and VC discussions on Twitter. "
These interpretations are not mutually exclusive, of course.
Essentially, the pattern "X Twitter" is roughly equivalent to "The X-o-sphere" and similar formulations WRT weblogs, but a bit more straightforward.
I tried with one comment and it was down voted to -1. Probably just one parent comment as prompt was not enough to contextualise it properly.
There are some GPT-3 generated comments in the big GPT-3 HN thread: