4 ms·

What's going on with this person? It looks like she really exists, her LinkedIn shows a solid network and the links seem to check out, and the Twitter account e

by tinco 5y ago

What's going on with this person? It looks like she really exists, her LinkedIn shows a solid network and the links seem to check out, and the Twitter account exists since 2009. But everything about her presence online seems fabricated, is she paying a service to generate these for her? What's the game, does this give her clout in the insurance industry somehow? The generated tweets are nonsense but seem to be fairly high quality nonsense. And the replies to Paul's reply were surreal, a bunch of them were just people having fun but a bunch looked like they were auto generated as well. I feel at some point big machine learning networks will interact with humans so much through Twitter and other networks that eventually they will learn how to sway and manipulate people. I think it's one of the ultimate weapons a super power could create using machine learning. Of course looking at this it seems they'd likely get stuck just talking nonsense to their competitors.

the__alchemist 5y ago
Re your last paragraph: Neal Stephenson explores this in his book 'Fall; or, Dodge in Hell'.
- robbedpeter 5y ago
  APEs - autonomous proxies for execration. Social media bots designed to offer up plausible commentary ranging from accurate to bizarrely disconnected from reality, deployed at scale in order to overwhelm the information available on a person such that discerning reality becomes impossible. You could probably do it with gpt-2, but gpt-neo's smallest model is perfectly sufficient to this task. All you need is a botnet and a sufficient number of nodes that can get away with a ~1.5gb download, then running a ~600mb model in memory. And antiCaptcha and other logistics payload. (Now, draw an owl.) Previous comment on hn: https://news.ycombinator.com/item?id=21977216 https://news.ycombinator.com/item?id=21977216 https://en.wikipedia.org/wiki/Fall;_or,_Dodge_in_Hell https://en.wikipedia.org/wiki/Fall;_or,_Dodge_in_Hell Twitter style social media is susceptible to this, dall-e and lithe synthetic media could make fb and ig plausible posts. Not a bad response to outrage mobbing,imo.
simonsarris 5y ago
I have a feeling we will see a future like this, in some places, for some topics. Writing about GPT-3 I mentioned: > How do you get to the point, then, where you're only filtering by human comments? What happens when its 99,000 comments, and 99% of them bots?? Will you know? Will you be influenced? > Will the average person be influenced? If advertising and the idea of saying something over and over works, and words are free, and you can GPT-3 your hand-rolled (or state authorized) propaganda every day in every comment thread in every forum to look like an average person is advocating for X, will other average people be more convinced of X? https://simonsarris.com/commentdammerung https://simonsarris.com/commentdammerung
- pphysch 5y ago
  Most people are already getting a fully synthetic worldview provided to them by powerful media companies. GPT-3/etc further saturates that space but doesn't add anything radically new to the picture.
  - rg111 5y ago
    It does add scale to the picture. And yes, it is radically impactful.
    - pphysch 5y ago
      Scaling a wetware troll farm of 10 trolls to a bot farm of 1000 bots isn't radically impactful unless you are competing with other troll/bot farms. The people will get highly synthetic information either way. The powerful will continue to use the most effective influence mediums available to them. As it was.
      - rg111 5y ago
        In my experience, it doesn't work like that. Some common talking points by either side of the American political spectrum gets regurgitated and propagated a lot, and by true believers. Also by troll farms. But those are not what I am worried about. I am worried about some issues and opinions getting dragged from the fringes to the mainstream by sheer scale. There are some cases which are not widely talked about- let's say it has a hundred deluded supporters. Now, a corporation might hire a high-tech troll farm that produces GPT-like-tech enabled troll accounts running wild in platforms and commenting wild stuff. There can be an effect in the minds of the undecided, or even people on the other side- "if so many people are saying this..." Thus is begins. Decisions can be swayed and opinions can be manufactured this way.
emerged 5y ago
I was honestly under the impression that machine learning networks are already doing exactly that. But for sure they will continue to get better and better at it and become much more convincing.
rg111 5y ago
> I feel at some point big machine learning networks will interact with humans so much through Twitter and other networks that eventually they will learn how to sway and manipulate people. It has already happened.

by tinco 5y ago

the__alchemist 5y ago

Re your last paragraph: Neal Stephenson explores this in his book 'Fall; or, Dodge in Hell'.

robbedpeter 5y ago
APEs - autonomous proxies for execration. Social media bots designed to offer up plausible commentary ranging from accurate to bizarrely disconnected from reality, deployed at scale in order to overwhelm the information available on a person such that discerning reality becomes impossible. You could probably do it with gpt-2, but gpt-neo's smallest model is perfectly sufficient to this task. All you need is a botnet and a sufficient number of nodes that can get away with a ~1.5gb download, then running a ~600mb model in memory. And antiCaptcha and other logistics payload. (Now, draw an owl.) Previous comment on hn: https://news.ycombinator.com/item?id=21977216 https://news.ycombinator.com/item?id=21977216 https://en.wikipedia.org/wiki/Fall;_or,_Dodge_in_Hell https://en.wikipedia.org/wiki/Fall;_or,_Dodge_in_Hell Twitter style social media is susceptible to this, dall-e and lithe synthetic media could make fb and ig plausible posts. Not a bad response to outrage mobbing,imo.

simonsarris 5y ago

I have a feeling we will see a future like this, in some places, for some topics. Writing about GPT-3 I mentioned: > How do you get to the point, then, where you're only filtering by human comments? What happens when its 99,000 comments, and 99% of them bots?? Will you know? Will you be influenced? > Will the average person be influenced? If advertising and the idea of saying something over and over works, and words are free, and you can GPT-3 your hand-rolled (or state authorized) propaganda every day in every comment thread in every forum to look like an average person is advocating for X, will other average people be more convinced of X? https://simonsarris.com/commentdammerung https://simonsarris.com/commentdammerung

pphysch 5y ago
Most people are already getting a fully synthetic worldview provided to them by powerful media companies. GPT-3/etc further saturates that space but doesn't add anything radically new to the picture.
- rg111 5y ago
  It does add scale to the picture. And yes, it is radically impactful.
  - pphysch 5y ago
    Scaling a wetware troll farm of 10 trolls to a bot farm of 1000 bots isn't radically impactful unless you are competing with other troll/bot farms. The people will get highly synthetic information either way. The powerful will continue to use the most effective influence mediums available to them. As it was.
    - rg111 5y ago
      In my experience, it doesn't work like that. Some common talking points by either side of the American political spectrum gets regurgitated and propagated a lot, and by true believers. Also by troll farms. But those are not what I am worried about. I am worried about some issues and opinions getting dragged from the fringes to the mainstream by sheer scale. There are some cases which are not widely talked about- let's say it has a hundred deluded supporters. Now, a corporation might hire a high-tech troll farm that produces GPT-like-tech enabled troll accounts running wild in platforms and commenting wild stuff. There can be an effect in the minds of the undecided, or even people on the other side- "if so many people are saying this..." Thus is begins. Decisions can be swayed and opinions can be manufactured this way.

emerged 5y ago

I was honestly under the impression that machine learning networks are already doing exactly that. But for sure they will continue to get better and better at it and become much more convincing.

rg111 5y ago

> I feel at some point big machine learning networks will interact with humans so much through Twitter and other networks that eventually they will learn how to sway and manipulate people. It has already happened.