4 ms·

The AI could comment as well as a human until it gets to its payload, which is a scam. People will vote it down after they get taken in by the scam, which is t

by Jiro 4y ago

The AI could comment as well as a human until it gets to its payload, which is a scam. People will vote it down after they get taken in by the scam, which is too late. Or maybe not even then, if the payload isn't a scam per se, but something that's deliberately designed by the AI's developer to be slanted. A million AIs that are indistinguishable from humans but occasionally throw in a comment about how they got sick from Pepsi could make things very profitable for the Coca-Cola company. Also, upvotes are an imperfect measure of how good the post is and AIs would be able to game the measure.

ShredKazoo 4y ago
>People will vote it down after they get taken in by the scam, which is too late. Not too late to create training data by which the social media platform can detect and remove subsequent scams. Especially if they 'report' rather than downvote. >A million AIs that are indistinguishable from humans but occasionally throw in a comment about how they got sick from Pepsi could make things very profitable for the Coca-Cola company. How is this different from Coca-Cola paying influencers to post negative stuff about Pepsi? >Also, upvotes are an imperfect measure of how good the post is and AIs would be able to game the measure. How is this different from humans gaming the measure?
- czzr 4y ago
  The ability to generate nonsense/noise at immense scale for essentially zero cost.
- Jiro 4y ago
  >Not too late to create training data by which the social media platform can detect and remove subsequent scams. How is that going to happen? We're assuming the AI posts exactly like a human until the comment about Pepsi comes out. It'll be impossible to distinguish from a human, by assumption. Or are you going to just have the site reject everything which mentions Pepsi negatively? >How is this different from Coca-Cola paying influencers to post negative stuff about Pepsi? AIs don't need to be paid.
  - ShredKazoo 4y ago
    >How is that going to happen? We're assuming the AI posts exactly like a human until the comment about Pepsi comes out. It'll be impossible to distinguish from a human, by assumption. Or are you going to just have the site reject everything which mentions Pepsi negatively? You could have a reporting system. >AIs don't need to be paid. At scale, this sort of service is not going to be free. But I suppose AI could make it a lot cheaper.
- yunwal 4y ago
  > How is this different from Coca-Cola paying influencers to post negative stuff about Pepsi? Coca cola can't have millions of influencers on their payroll (or at least, it wouldn't be economically profitable for them to do that). An AI could easily create and post on accounts on a scale that dwarfs real human-generated content.
  - ShredKazoo 4y ago
    80/20 rule -- 20% of influencers generate 80% of the views. Don't need to pay everyone. https://www.ftc.gov/business-guidance/resources/disclosures-101-social-media-influencers https://www.ftc.gov/business-guidance/resources/disclosures-... ^ Wouldn't this apply to ChatGPT bots as well?

by Jiro 4y ago

ShredKazoo 4y ago

>People will vote it down after they get taken in by the scam, which is too late. Not too late to create training data by which the social media platform can detect and remove subsequent scams. Especially if they 'report' rather than downvote. >A million AIs that are indistinguishable from humans but occasionally throw in a comment about how they got sick from Pepsi could make things very profitable for the Coca-Cola company. How is this different from Coca-Cola paying influencers to post negative stuff about Pepsi? >Also, upvotes are an imperfect measure of how good the post is and AIs would be able to game the measure. How is this different from humans gaming the measure?

czzr 4y ago
The ability to generate nonsense/noise at immense scale for essentially zero cost.
Jiro 4y ago
>Not too late to create training data by which the social media platform can detect and remove subsequent scams. How is that going to happen? We're assuming the AI posts exactly like a human until the comment about Pepsi comes out. It'll be impossible to distinguish from a human, by assumption. Or are you going to just have the site reject everything which mentions Pepsi negatively? >How is this different from Coca-Cola paying influencers to post negative stuff about Pepsi? AIs don't need to be paid.
- ShredKazoo 4y ago
  >How is that going to happen? We're assuming the AI posts exactly like a human until the comment about Pepsi comes out. It'll be impossible to distinguish from a human, by assumption. Or are you going to just have the site reject everything which mentions Pepsi negatively? You could have a reporting system. >AIs don't need to be paid. At scale, this sort of service is not going to be free. But I suppose AI could make it a lot cheaper.
yunwal 4y ago
> How is this different from Coca-Cola paying influencers to post negative stuff about Pepsi? Coca cola can't have millions of influencers on their payroll (or at least, it wouldn't be economically profitable for them to do that). An AI could easily create and post on accounts on a scale that dwarfs real human-generated content.
- ShredKazoo 4y ago
  80/20 rule -- 20% of influencers generate 80% of the views. Don't need to pay everyone. https://www.ftc.gov/business-guidance/resources/disclosures-101-social-media-influencers https://www.ftc.gov/business-guidance/resources/disclosures-... ^ Wouldn't this apply to ChatGPT bots as well?