4 ms·

Hello from an AI safety ninny. I have posted these two concerns multiple times and no one posted any counters to them. 1. There was https://www.youtube.com/wat

by _ugfj 3y ago

Hello from an AI safety ninny. I have posted these two concerns multiple times and no one posted any counters to them. 1. There was https://www.youtube.com/watch?v=xoVJKj8lcNQ https://www.youtube.com/watch?v=xoVJKj8lcNQ where they argued for 2028 and on will be AI elections where the person with most computing power wins. 2. Propaganda produced by humans on small scale killed 300 000 people in the US alone in this pandemic https://www.npr.org/sections/health-shots/2022/05/13/1098071284/this-is-how-many-lives-could-have-been-saved-with-covid-vaccinations-in-each-sta https://www.npr.org/sections/health-shots/2022/05/13/1098071... imagine the next pandemic when it'll be produced on an industrial scale by LLMs. Literally millions will die of it.

root_axis 3y ago
None of this seems related to LLMs. Propaganda produced by humans is effective because of the massive scale of distribution, being able to produce more variations of the same talking points doesn't change the threat risk.
- semi 3y ago
  Being able to produce more variations of the same talking points sounds really useful for increasing the scale of distribution - you can much more easily maintain more legitimate looking sock puppet accounts that can appear to more organically agree with your talking points.
  - root_axis 3y ago
    I don't think it moves the needle much at all. At the end of the day the scaling bottleneck is access to gullible or ideologically motivated eyeballs. The internet is already over-saturated with more propaganda than any individual can consume, adding more shit to the pile isn't going to suddenly convince a reasonable person that vaccines have microchips inside.
  - gmerc 3y ago
    Have you seen Fox News?
hatenberg 3y ago
The fix to neither lies in technology. And it doesn't lie in AI alignment. We cannot align AI because WE are not aligned. For 50% of congress (you can pick your party as the other side, regardless which one you are), the "AI creates misinformation" narrative sounds like "Oh great, I get re-elected easier"). This is a governance and regulation problem - not a technology problem. Big tech would love you to think that "they can solve AI" if we follow the China model of just forcing everything to go through big tech and they'll regulate it pliantly in exchange for market protection and the more pressure there is on their existing growth models, the more excited they are about pushing this angle. Capitalism requires constant growth, which unfortunately is very challenging given diminishing returns in R&D. You can only optimize the internal combusion engine for so long before the costs of incremental increases start killing your profit, and the same is true to any other technology. And so now we have big Knife Company who are telling governments that they will only sell blunt knifes and nobody will ever get hurt, and that's the only way nobody gets hurt because if there's dozens of knife stores, who is gonna regulate those effectively. So no, I don't think your concerns are actually related to AI. They are related to society, and you're buying into the narrative that we can fix it with technology if only we give the power over that technology to permanent large gate-keepers. The risks you flag are related to: - Distribution of content at scale. - Erosion of trust (anyone can buy a safety mark). - Lack of regulation and enforcement of said risks. - The dilemma of where the limits of free speech and tolerance lie. Many of those have existed since Fox News.
- chx 3y ago
  What you are saying is neither here, nor there. All we need to do is ban generative AIs. Now. Before it's too late. Simple.
  - gmerc 3y ago
    “All we have to do is ban gunpowder and our castles will protect us” “All we have to do is prohibit alcohol” “All we have to do is prevent printing press ownership” You cannot be that naive AI actually is even simpler than these technologies - the math is already out, the GPUs powering every video game. That train left the station somewhere around Data is all you need. You are clinging to the illusion that humans can ban technology that is power relevant.
pk-protect-ai 3y ago
You should not worry about AI problems by 2028. Dozens of millions worldwide will die from climate-related problems by that time. Literally, nobody will care about the topic of AGI anymore.
- adroniser 3y ago
  You should worry about both problems. You're telling me that AI isn't going to improve it's video capabilities in the next 4 years enough to make convincing deepfakes?
  - pk-protect-ai 3y ago
    It already does. And I'm not worried. This is to be mitigated by law enforcement not by AI forbidding.
    - adroniser 3y ago
      How can you effectively enforce anything if the models are open source? How do you draw the line if a deepfake is not defamatory (making someone say something they didn't say) but in fact just makes someone look silly https://en.wikipedia.org/wiki/Ed_Miliband_bacon_sandwich_photograph https://en.wikipedia.org/wiki/Ed_Miliband_bacon_sandwich_pho.... Or using LLMs to scale up what happened with cambridge analytica and create individualized campaigns and bots to influence elections?
      - pk-protect-ai 3y ago
        You should handle it as any other crime. Why do you ask? It does not matter how good the gun is, what matters is who has pulled the trigger.
        adroniser 3y ago
        Yes but if we had the ability to download a gun from the internet anonymously with no way to feasibly get the identity of the person downloading the gun I think we would be right to be concerned. Especially if you could then shoot that gun at someone anonymously.

by _ugfj 3y ago

root_axis 3y ago

None of this seems related to LLMs. Propaganda produced by humans is effective because of the massive scale of distribution, being able to produce more variations of the same talking points doesn't change the threat risk.

semi 3y ago
Being able to produce more variations of the same talking points sounds really useful for increasing the scale of distribution - you can much more easily maintain more legitimate looking sock puppet accounts that can appear to more organically agree with your talking points.
- root_axis 3y ago
  I don't think it moves the needle much at all. At the end of the day the scaling bottleneck is access to gullible or ideologically motivated eyeballs. The internet is already over-saturated with more propaganda than any individual can consume, adding more shit to the pile isn't going to suddenly convince a reasonable person that vaccines have microchips inside.
- gmerc 3y ago
  Have you seen Fox News?

hatenberg 3y ago

The fix to neither lies in technology. And it doesn't lie in AI alignment. We cannot align AI because WE are not aligned. For 50% of congress (you can pick your party as the other side, regardless which one you are), the "AI creates misinformation" narrative sounds like "Oh great, I get re-elected easier"). This is a governance and regulation problem - not a technology problem. Big tech would love you to think that "they can solve AI" if we follow the China model of just forcing everything to go through big tech and they'll regulate it pliantly in exchange for market protection and the more pressure there is on their existing growth models, the more excited they are about pushing this angle. Capitalism requires constant growth, which unfortunately is very challenging given diminishing returns in R&D. You can only optimize the internal combusion engine for so long before the costs of incremental increases start killing your profit, and the same is true to any other technology. And so now we have big Knife Company who are telling governments that they will only sell blunt knifes and nobody will ever get hurt, and that's the only way nobody gets hurt because if there's dozens of knife stores, who is gonna regulate those effectively. So no, I don't think your concerns are actually related to AI. They are related to society, and you're buying into the narrative that we can fix it with technology if only we give the power over that technology to permanent large gate-keepers. The risks you flag are related to: - Distribution of content at scale. - Erosion of trust (anyone can buy a safety mark). - Lack of regulation and enforcement of said risks. - The dilemma of where the limits of free speech and tolerance lie. Many of those have existed since Fox News.

chx 3y ago
What you are saying is neither here, nor there. All we need to do is ban generative AIs. Now. Before it's too late. Simple.
- gmerc 3y ago
  “All we have to do is ban gunpowder and our castles will protect us” “All we have to do is prohibit alcohol” “All we have to do is prevent printing press ownership” You cannot be that naive AI actually is even simpler than these technologies - the math is already out, the GPUs powering every video game. That train left the station somewhere around Data is all you need. You are clinging to the illusion that humans can ban technology that is power relevant.

pk-protect-ai 3y ago

You should not worry about AI problems by 2028. Dozens of millions worldwide will die from climate-related problems by that time. Literally, nobody will care about the topic of AGI anymore.

adroniser 3y ago
You should worry about both problems. You're telling me that AI isn't going to improve it's video capabilities in the next 4 years enough to make convincing deepfakes?
- pk-protect-ai 3y ago
  It already does. And I'm not worried. This is to be mitigated by law enforcement not by AI forbidding.
  - adroniser 3y ago
    How can you effectively enforce anything if the models are open source? How do you draw the line if a deepfake is not defamatory (making someone say something they didn't say) but in fact just makes someone look silly https://en.wikipedia.org/wiki/Ed_Miliband_bacon_sandwich_photograph https://en.wikipedia.org/wiki/Ed_Miliband_bacon_sandwich_pho.... Or using LLMs to scale up what happened with cambridge analytica and create individualized campaigns and bots to influence elections?
    - pk-protect-ai 3y ago
      You should handle it as any other crime. Why do you ask? It does not matter how good the gun is, what matters is who has pulled the trigger.
      - adroniser 3y ago
        Yes but if we had the ability to download a gun from the internet anonymously with no way to feasibly get the identity of the person downloading the gun I think we would be right to be concerned. Especially if you could then shoot that gun at someone anonymously.