Hacker News Reader

Comment by toisanji | Hacker News Reader

I thought he was working on AGI? Gave up already?

I don't feel like this precludes him from working on AGI -- and for him, it seems to be related: https://www.facebook.com/permalink.php?story_fbid=2110408722...

The road to AGI is paved with yak shavings.

First I have to have a stable OS. Then I have to make kernel updates to support the kind of process management I want. Then I have to fix the other kernel bugs affecting my update. Then I have to modify the network drivers. Then I have to ... 5 years later what was I even working on in the first place?

To make AGI from scratch, first you have to create the universe.

Not the original source, but better? ;)

https://www.youtube.com/watch?v=zSgiXGELjbc

That's one of the funniest things I've read on HN.

But first we need to breed the yaks

And walls

My dream is to do what Palmer Luckey did - build a successful computer graphics-based startup and then lure John Carmack to be CTO.

I've got a snowball's chance in hell of being that successful, but I do envy the fact Palmer was able to hire him. What an incredible hire.

Then again, he did have Facebook money.

It's cool that he's got the autonomy to pursue AGI after guiding Oculus into a good position.

Oculus wouldn't have had that buzz without Carmack pushing the original Rift DIY kit before their Kickstarter. He was already commenting on Palmers DIY creations in the community where the ideas for Rift were born.

It was more or less a community project, because without community input in those forum threads Palmer woulnd't even had considered glasses, he was deciding between VR glasses and a 3D monitor in his first posts there, and got nudged by a member towards glasses, lol. He just never stopped absorbing information and kept iterating on his franken HMDs and prototypes, and one day when he had already reached a good level Carmack showed up.

At this point in Carmack's career, I doubt money is a top three reason for why he takes on a project.

Maybe he is testing AGI by seeing if it can pass the turing test for writing device drivers.

Most people who actually work in AI know that AGI being possible is still an IF and IF it is possible, it is centuries or even further away (basically some indeterminate time frame because we don't even know if we're on the right path). The techniques we're using today are likely not going to be applicable to AGI. John Carmack is very smart, but I wouldn't place him on the same pedestal as any of our great historical polymaths and IMO if a single person is going to make any progress in AGI it's going to be someone like Leibniz or Euler.

That's an interesting perspective: why do you believe that it will be an individual who will make a breakthrough, as opposed to a web of Carmack-like folks?

I actually don't believe an individual will make a breakthrough, that was more of a if an individual did do it statement. IMO it's actually going to be the result of a lot of methodical work by an army of researchers over a really long period of time.

> IF it is possible, it is centuries or even further away

Just out of curiosity, would you have been willing to bet that something most people will agree on being AGI will not be developed before 2221 at the earliest?

Because given the history of science and technology since 1821, that seems like a statement that's just drawn out of thin air. "I don't know" would make sense. "Definitely not in the next 200 years" does not make much sense.

Sorry, the time frame was arbitrary, I clarified in parens what I intended by that statement (an indeterminate amount of time)

Not to be judgmental, and to support your argument, but two points: I don't think we know how smart he is and I don't think it's that important. Euler was to me a balanced person as opposed to a extreme person like Grothendieck. So, to each their own.

Second point: I think almost everyone being promoted for being involved in AI or AGI are probably irrelevant in the longer run. Self driving cars are at this point something of a social problem rather than a clear mathematical problem. So, whether we solve the current list of business related AI problems is probably immaterial for mathematics, but important for public opinion. I haven't read or seen any projects that are even remotely close to AGI, unless someone is hiding something.

Whether some person x is doing (edit: working on) AGI I guess matters only in the social or business sense.

I am not saying that he will solve AGI on his own (he did quit rocket science after all), but you are greatly underestimating the impact he can have in the field. He basically evolved almost on his own two multi-billion dollar markets by doing noticeable technology leaps and finishing things at the right time. That must count for something. I would bet he cannot solve AGI, but he will surely leave something inspiring behind.

> He basically evolved almost on his own two multi-billion dollar markets

Not sure what this had to do with AGI, I don't even think there is a relationship in the skillset required

There is basically none. People on HN have this weird god-worship of John Carmack. He's a great engineer and businessman, but a top-tier theoretical researcher? He has less to offer than most PhD students in these fields. It's kind of insane the cult following he has.

There’s no reason AI can’t happen from a single jump in innovation

> [know that AGI] is centuries or even further away (basically some indeterminate time frame because we don't even know if we're on the right path).

How can you, in the same sentence, say the time frame is indeterminate and that it is 100+ years away?

We just don't know, period.

Well what we do know is that current AI methods are very task specific and there is nothing that generalizes well to many tasks. If it was happening in the next century we would at least have some idea how to start generalizing to many tasks with a single model or something comparable to that. We had the analytical engine in 1830s and we didn't see a turing machine or lambda calculus until the 1930s

It is extremely hard (read impossible) to predict what is going to evolve exponentially or asymptomatically in the future. Look at computing and aerospace for example. Take the state of those 2 domains and ask someone in 1900 where they think we would be in 2000, I am pretty sure they would have been quite far off.

Working at an AI lab myself, while I agree with your statement about the state of AI today, I don't know any of my colleague that would be confident to say it's 100+ years away.

> If it was happening in the next century we would at least have some idea

Just like at the beginning of the 20th century we knew ~0 about quantum physics and less than 50 years later A-bombs were getting dropped.

Or how, at the same time no plane could achieve sustained/controlled flight (and the goal was merely to stay above ground) and less than 70 years later not only were we leaving Earth's atmosphere: a man was walking on the moon.

We have no idea how long it will take to achieve AGI, let's leave it at that.

I kind of get the feeling, that we're trying to build a plane by making the wings flap like a bird.

Reply on news.ycombinator.com