3 ms·

Wait, is this supposed to have already happened, in your view? Shouldn't people be reading generated novels, teaching from generated textbooks, optioning genera

by pvg 2y ago

Wait, is this supposed to have already happened, in your view? Shouldn't people be reading generated novels, teaching from generated textbooks, optioning generated movie scripts, etc if that were the case?

triyambakam 2y ago
It's an interesting and maybe useful idea definitely, but I agree I don't think it's passed yet.
- nonrandomstring 2y ago
  I think we already passed the point because it's experimentally undecidable now. "distinguishing between human and AI written text or generated images was no longer easy or even possible." I mean, how would you know? This seems to set up an falsifiable test. Unless you've strict experimental conditions in which all "AI" material is verified and labelled as such? The reality is that "AI" is a tool of deception. Writers who use it disguise their source and try to pass generative work off as their own. Publishers make money by pretending to have real writers. "AI" itself operates as "artifice", as "fake people". The conditions for discrimination aren't isolated enough, it's already deeply entwined with capitalist/political values. And further, there is an ideological bias overlaid on the problem. Some proponents or detractors simply won't accept known results even where there's a good proof or chain of evidence.
low_tech_love 2y ago
Are we not? I am pretty sure the vast majority of things written/published in the past couple of years have had a significant amount of AI in it.
- pvg 2y ago
  I am pretty sure Ok but I'm not asking about "the vast majority of writing" because that's not very meaningful - we do a lot of rote, predictable, functional writing. What's the evidence for the kind of writing I asked about? Why can't I ask ChatGPT to write me an award-winning novel? Or a middling, publishable novel.
nirfyi 2y ago
Yes it already happend. Right now if I give you a crafted article with some fine tuning you won't be able to tell the difference. Just the openai flavours give clear signals but others might not.

by pvg 2y ago

triyambakam 2y ago

It's an interesting and maybe useful idea definitely, but I agree I don't think it's passed yet.

nonrandomstring 2y ago
I think we already passed the point because it's experimentally undecidable now. "distinguishing between human and AI written text or generated images was no longer easy or even possible." I mean, how would you know? This seems to set up an falsifiable test. Unless you've strict experimental conditions in which all "AI" material is verified and labelled as such? The reality is that "AI" is a tool of deception. Writers who use it disguise their source and try to pass generative work off as their own. Publishers make money by pretending to have real writers. "AI" itself operates as "artifice", as "fake people". The conditions for discrimination aren't isolated enough, it's already deeply entwined with capitalist/political values. And further, there is an ideological bias overlaid on the problem. Some proponents or detractors simply won't accept known results even where there's a good proof or chain of evidence.

low_tech_love 2y ago

Are we not? I am pretty sure the vast majority of things written/published in the past couple of years have had a significant amount of AI in it.

pvg 2y ago
I am pretty sure Ok but I'm not asking about "the vast majority of writing" because that's not very meaningful - we do a lot of rote, predictable, functional writing. What's the evidence for the kind of writing I asked about? Why can't I ask ChatGPT to write me an award-winning novel? Or a middling, publishable novel.

nirfyi 2y ago

Yes it already happend. Right now if I give you a crafted article with some fine tuning you won't be able to tell the difference. Just the openai flavours give clear signals but others might not.