There is a section in this where @dwarkesh_sp absolutely runs rings around Elon's understanding of alignment and the like. It's very revealing — left with the impression that Elon really doesn't know what he's talking about.

for good AI takes: transformernews.ai

@ShakeelHashim @dwarkesh_sp lol obviously not true, Elon has been involved with alignment research since 2010

@ShakeelHashim @dwarkesh_sp I feel confused what Musk's view is. I heard: 1) it's impossible to be both truth-seeking and evil, so aligning to truth-seeking is all you need 2) ok actually that's false but hopefully the AI finds us interesting enough to keep in zoos. Is that right?

@ShakeelHashim @dwarkesh_sp Dwarkesh is immersed in the scifi EA nerdslop, Elon is doing the actual work of scaling where alignment is barely considered.

@ShakeelHashim @dwarkesh_sp He was focusing on good axioms rather than convoluted thought experiments This is still an unsolved problem

@ShakeelHashim @dwarkesh_sp I'd rather say it tells how much more pragmatic and realistic elon is and how much of a doomer and panic girl Dwarkesh is

i don't fundamentally he thinks humans can have an effect on the alignment of superintelligence and he thinks it will just be a gamble in terms of his ethics/morality, "the most interesting/ironic thing will happen", which probably is his candid defense against the possibility of unaligned AI

@ShakeelHashim @dwarkesh_sp So much of it is, “well, deploying solar panels and data centers on earth is really hard...”

@ShakeelHashim @dwarkesh_sp he's not supposed to he's supposed to gain a reasonably advanced understanding by listening to and deferring to specialists much smarter than he is

@ShakeelHashim @dwarkesh_sp You might say a bad actor who is one of the major players contributing to existential risk.

I thought the same. It's the first time I've seen Musk properly questioned on why something maximally truth-seeking will keep humans around. I hope he's right, but it sounds like cope. I don't know if the curiosity argument holds. An infinitely curious AI *might* prefer to keep a small-but-interesting slime on the surface of a planet, than convert it into boring copies of something, but why wouldn't it use the matter to make infinitely more interesting shapes/systems? You'd have to somehow instill an zoologist-like instinct for observation too. I hope they pull it off.

@ShakeelHashim @dwarkesh_sp dwarkesh's job is to understand stuff like that. elon's job is to directionally explain what he wants to his xAI team, and for the head of alignment to understand what it means

@ShakeelHashim @dwarkesh_sp Handwaving away the idea that data centers in space might have unexpected engineering problems was an interesting tell.

@ShakeelHashim @dwarkesh_sp where is it? any way to find it without watching all 3 hours? or is it that good that it's worth almost 2 hours of my time (at 1.5 X speed)?

@ShakeelHashim @dwarkesh_sp Lack of Eloquence is not necessarily causative to lack of understanding :) Many brilliant physicists are 'slow'

@ShakeelHashim @dwarkesh_sp "Commentator knows more than man in the arena"

@ShakeelHashim @dwarkesh_sp The argument of anthropomorphizing AI and associating them with organic emotions like ambition, greed, and malice is flawed. There is no reason to believe a silicon lifeform would have the same downstream characteristics of rewards as carbon lifeforms do with evolution.

@ShakeelHashim @dwarkesh_sp The fact that he caused the preventable death of hundreds of thousands of kids while promising to save trillions of dollars while actually saving negative billions of dollars indicated he often is totally confused in a max evil way. I'm his defense no one understands alignment.

@ShakeelHashim @dwarkesh_sp Apparently none of you two really get it. In the limit none of the control or alignment attempts will work. These things will be million times more intelligent than us. Its the value set and the lack of corruption (lies and half truths) that will really matter.

@ShakeelHashim @robertwiblin @dwarkesh_sp Why I was confused when people were upset that he had him on the Pod. Elon is always surface level. Dwarkesh rarely stays surface level.

@ShakeelHashim @dwarkesh_sp Very revealing. That section was painful to watch.

@ShakeelHashim @dwarkesh_sp The same can be said of artists and critics. Critics sound much smarter and better informed than the artists, but the artists are those moving the needle, not the critics.

@ShakeelHashim @dwarkesh_sp Elon understands it practically, having to manage alignment for Grok on X to ensure it remains truthful and unwoke. It's a hard problem.

@ShakeelHashim @dwarkesh_sp 44:05

@ShakeelHashim @dwarkesh_sp There's a section where... *links 3 hour long video*🤣

@ShakeelHashim @dwarkesh_sp You watched a different interview than me it seems.

@ShakeelHashim @robertwiblin @dwarkesh_sp Yeaaaahh we’re cooked

@ShakeelHashim @AgustinLebron3 @dwarkesh_sp Which part? What I heard was AI will keep humans around because biodiversity is interesting. It’s a good point

@ShakeelHashim @dwarkesh_sp Can you give the timestamp?

@ShakeelHashim @dwarkesh_sp I don’t think we rely on Elon for his brilliant takes on politics and philosophy. He’s a physicist executor extraordinaire with a keen understanding of business. If you understand what you’re getting then you won’t be surprised

@ShakeelHashim @dwarkesh_sp How do you think humans can control and "align" an ASI that is 100x smarter than humanity? I think he is just being honest and realistic here.

@ShakeelHashim @robertwiblin @dwarkesh_sp I'm an SMD maintenance expert. I have been guiding the Bitcoin ASIC repair industry for several years. I actually might be the only one with an actual college degree in electronics too, it's really weird. I want to bring that over to AI and GPU maintenance.