Eliezer Yudkowsky: Dangers of AI and the End of Human Civilization | Lex Fridman Podcast #368

There's no real acknowledgement in this conversation of the existential threat. I would recommend listening to Eliezer Yudkowsky on the subject [1] [2]: > If somebody builds a too-powerful AI, under present conditions, I expect that every single member of the human species and all biological life on Earth dies shortly thereafter. > There’s no proposed plan for how we could do any such thing and survive. OpenAI’s openly declared intention is to make some future AI do our AI alignment homework. Just hearing that this is the plan ought to be enough to get any sensible person to panic. The other leading AI lab, DeepMind, has no plan at all.There’s no proposed plan for how we could do any such thing and survive. OpenAI’s openly declared intention is to make some future AI do our AI alignment homework. Just hearing that this is the plan ought to be enough to get any sensible person to panic. The other leading AI lab, DeepMind, has no plan at all. It is a valid criticism Yudkowsky and others are over-staging the abilities of current day GPT. For example Eliezer says "can it reason? It can play chess". But I looked at a GPT chess game (actually published in an article on his less wrong site). GPT has no idea how to play chess. It can make some moves in the opening that match opening theory which it has seen, etc. But then it will lose a piece because it has no actual ability to think 2 moves ahead. And then when it gets into the end game it tries to make illegal king moves. As impres

No I don't think so. Not because of the content but rather the lack of it. The people that already have some idea of the missing content are not the general audience that would benefit from this. In my opinion a TED talk is not an appropriate format. TED barely gave him enough time to tell us the sky is going to be falling. People will need more depth to see what that might look like and how it might apply to them and what it even means to look for it. I believe the format should be that of a 3 or 4 hour long-format podcast. Oh and not just one but many. There should be at most 3 guests on each podcast or it turns into a disruptive debate but each guest should be very intelligent and should help Eliezer go deeper into each rabbit hole and illuminate the path the AGI may take to reach it's desired outcome and what logic it would have used to reach this outcome. There should also be discussions about how to mitigate the highest risks without entirely banning the technology since it is impossible to ban a framework or toolset world-wide. People should walk away with an understanding of how bad things may/will happen and how they might contribute to mitigating said bad things in their career field. Or at least spot when they are their colleagues may be empowering automation to do bad things . There should also be a 3 hour long podcast that stays at surface level but also helps people especially legislators, regulators and military intelligence understand the topic enough to make

tl;dr it's the same feature he's talking about, but also he's wrong in a very Yudkowsky way. I was curious about this so I hunted it down: [this video] He speculates that Wikipedia got the idea from Arbital. Insofar as I can tell, that seems to be a wiki-like website that started around 2015 -- I had never heard of it, so it might be older? That does predate this specific implementation of popups on Wikipedia, which is from 2018. The fun context here is actually discussed in the article -- the genesis of the Wikipedia popups was a community gadget on Wikipedia (Navigation Popups) that's from much earlier. Oldest version I could casually dig up on Wikipedia is from 2007. Unless Arbital is vastly older than I think (older than LessWrong itself, which is from 2009), the chain of inspiration is presumably in the other direction. Or completely independent. The article is, overall, about why it took so long to go from that working example to something more robust.