4 ms·

I appreciate that people are thinking about these things, but I still can't take the idea seriously that transformers represent a threat to democracy. Maybe wit

I appreciate that people are thinking about these things, but I still can't take the idea seriously that transformers represent a threat to democracy. Maybe with a massive enough supercomputer a country could run an AI IDE capable of end-to-end writing a device driver in Rust - but even that's not a given. Certainly, it's almost meaningless in the face of building our lives around a network of personal surveillance devices that we literally never part with. I'm just saying... ChatGPT is the least of our problems.

jasonfarnon 1y ago
Well do you think that social media could be a threat to democracy, as many have asserted, e.g., cambridge analytica, the US disinformation bureau etc? (Likewise that social media could be a threat to autocracy?) You don't see LLMs' impact on culture and society being at least as broad and thorough?
- root_axis 1y ago
  > Well do you think that social media could be a threat to democracy Yes, I certainly agree that it is and recognize all your examples. > You don't see LLMs' impact on culture and society being at least as broad and thorough? In the sense I think you're implying, I see them as having almost zero impact. Just because more crap is generated doesn't mean it's going to be more believable, or that it will even be seen. How many tokens do you think it would take a SOTA model to convince you that the earth is flat or that the moon-landing was a hoax? Do you think Trump supporters will start voting for Democrats if they see 100 anti-Trump posts in every comment section on the internet? The LLM isn't going to generate anything that we haven't all heard already.
  - TheOtherHobbes 1y ago
    With enough money, you absolutely could persuade Trump voters to vote for Democrats. This isn't up for debate. PR and advertising exist because this is absolutely possible, and has been for a long time. AI makes it much, much easier and more cost-effective. This was how Cambridge Analytica and Facebook swung the Brexit vote. They didn't send out blanket "EU bad, vote Brexit" content. They created targeted ads that addressed hot-button fear points for individuals and niche demographics and used Facebook's ad targetting system to deliver them. So some people were concerned about money for the UK's health system. They saw ads promising that Brexit would mean more money for the NHS. Others were concerned about potholes, or education, or - often - immigration. Every group saw ads that triggered their personal hot-button issues and persuaded them to vote against their interests.
  - int_19h 1y ago
    LLMs allow for much subtler forms of manipulation. For example, those Trump supporters could see 100 pro-Trump posts that very much toe the party line on the surface while introducing subtle contrarian points. And it's not that the LLM is going to generate something that we haven't seen before. The point is that it can do that at scale far exceeding what you can do with mere human troll bots. With LLMs you can literally have a bot assigned to every single person in the country that feeds them agitprop tuned for maximum agreement by that person.
ehnto 1y ago
It might just be a need for some creativity but the risks are pretty clear. The main vector is something that has already played out, which is massive misinformation campaigns across the web. But I feel like the humans will win that one long term, as bots fill the public web with bitter political rhetoric I think people will retreat to less politicised private communities. I certainly have. Another angle is as you noted, we basically surrender all our private data to corporations. What if a reigning political party decides that they need to develop an anti-terror model, that scans all communication from all people for Nasty Terror Thoughts then flags them for detainment. If the System has decided you are evil and the System is considered super intelligent, who is allowed to second guess it. Maybe though, evil thoughts are just disagreement with the reigning political party.
- root_axis 1y ago
  > The main vector is something that has already played out, which is massive misinformation campaigns across the web. I wrote this in a few other places, but this is a long foregone state of affairs. People's attention spans are already fully saturated, bloating the internet with a bunch of variations of the same crap isn't going to do anything that isn't already happening today. I don't need to generate a hundred million times what I can simply post once to a hundred million people. > What if a reigning political party decides that they need to develop an anti-terror model, that scans all communication from all people for Nasty Terror Thoughts then flags them for detainment This is already possible and happening (e.g. CSAM scanning). The crux of my point is that LLMs really aren't that big of a deal compared to the panopticon society that we've already built. The agents of the authoritarian control platform aren't going to become 10x more spooky because they installed a language model plugin.

jasonfarnon 1y ago

Well do you think that social media could be a threat to democracy, as many have asserted, e.g., cambridge analytica, the US disinformation bureau etc? (Likewise that social media could be a threat to autocracy?) You don't see LLMs' impact on culture and society being at least as broad and thorough?

root_axis 1y ago
> Well do you think that social media could be a threat to democracy Yes, I certainly agree that it is and recognize all your examples. > You don't see LLMs' impact on culture and society being at least as broad and thorough? In the sense I think you're implying, I see them as having almost zero impact. Just because more crap is generated doesn't mean it's going to be more believable, or that it will even be seen. How many tokens do you think it would take a SOTA model to convince you that the earth is flat or that the moon-landing was a hoax? Do you think Trump supporters will start voting for Democrats if they see 100 anti-Trump posts in every comment section on the internet? The LLM isn't going to generate anything that we haven't all heard already.
- TheOtherHobbes 1y ago
  With enough money, you absolutely could persuade Trump voters to vote for Democrats. This isn't up for debate. PR and advertising exist because this is absolutely possible, and has been for a long time. AI makes it much, much easier and more cost-effective. This was how Cambridge Analytica and Facebook swung the Brexit vote. They didn't send out blanket "EU bad, vote Brexit" content. They created targeted ads that addressed hot-button fear points for individuals and niche demographics and used Facebook's ad targetting system to deliver them. So some people were concerned about money for the UK's health system. They saw ads promising that Brexit would mean more money for the NHS. Others were concerned about potholes, or education, or - often - immigration. Every group saw ads that triggered their personal hot-button issues and persuaded them to vote against their interests.
- int_19h 1y ago
  LLMs allow for much subtler forms of manipulation. For example, those Trump supporters could see 100 pro-Trump posts that very much toe the party line on the surface while introducing subtle contrarian points. And it's not that the LLM is going to generate something that we haven't seen before. The point is that it can do that at scale far exceeding what you can do with mere human troll bots. With LLMs you can literally have a bot assigned to every single person in the country that feeds them agitprop tuned for maximum agreement by that person.

ehnto 1y ago

It might just be a need for some creativity but the risks are pretty clear. The main vector is something that has already played out, which is massive misinformation campaigns across the web. But I feel like the humans will win that one long term, as bots fill the public web with bitter political rhetoric I think people will retreat to less politicised private communities. I certainly have. Another angle is as you noted, we basically surrender all our private data to corporations. What if a reigning political party decides that they need to develop an anti-terror model, that scans all communication from all people for Nasty Terror Thoughts then flags them for detainment. If the System has decided you are evil and the System is considered super intelligent, who is allowed to second guess it. Maybe though, evil thoughts are just disagreement with the reigning political party.

root_axis 1y ago
> The main vector is something that has already played out, which is massive misinformation campaigns across the web. I wrote this in a few other places, but this is a long foregone state of affairs. People's attention spans are already fully saturated, bloating the internet with a bunch of variations of the same crap isn't going to do anything that isn't already happening today. I don't need to generate a hundred million times what I can simply post once to a hundred million people. > What if a reigning political party decides that they need to develop an anti-terror model, that scans all communication from all people for Nasty Terror Thoughts then flags them for detainment This is already possible and happening (e.g. CSAM scanning). The crux of my point is that LLMs really aren't that big of a deal compared to the panopticon society that we've already built. The agents of the authoritarian control platform aren't going to become 10x more spooky because they installed a language model plugin.