Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

John Carmack's new AGI company, Keen Technologies, has raised a $20M round

twitter.com

Post reply on HN
Print view View original

391–400 of 643 posts

Re: John Carmack's new AGI company, Keen Technologies, has raised a $20M round

#391

Post by BLKNSLVR »

gcau wrote: post #322

I believe to create AI, we need to first simulate the universe. It's the only way that makes sense to me apart from some magical algorithm people think will be discovered. I'm doubtful we'll reach it in our lifetimes, true AI running on supercomputers, it seems like the final, end all mission. Like switching your Minecraft world from survival to creative mode.

I had a similarly nihilistic thought just a couple of days ago. It's like the conundrum among the lines of "a map of sufficient detail ends up being the same size as the terrain being mapped".

To know the variables required for "intelligence" we need to recreate the universe in which intelligence came to be.

Re: John Carmack's new AGI company, Keen Technologies, has raised a $20M round

#392

Post by bhdlr »

rebelos wrote: post #186
madrox wrote: post #66

Earlier quoted context omitted.

I got the same impression, and maybe it still is. You can still raise money for a retirement project if the goal of the money is to hire a staff. VC money isn't solely for young 20-something founders who want to live their job.

Carmack sounds like someone who lives his job, so I don't think age/life stage is a factor here.

> Carmack sounds like someone who lives his job, so I don't think age/life stage is a factor here.

He said on lex Fridman that he's going all in on it. He said he was tired of his 1 day a week consultancy at oculus and wanted to work full time on this.

Re: John Carmack's new AGI company, Keen Technologies, has raised a $20M round

#393

Post by godelski »

8f2ab37a-ed6c wrote: post #256
Jack000 wrote: post #116

I'm very optimistic for near-term AGI (10 years or less). Even just a few years ago most in the field would have said that it's an "unknown unknown", we didn't have the theory or the models, there was no path forward and so it was impossible to predict. Now we have a fairly concrete idea of what a potential AGI might look like - an RL agent that uses a large transformer. The issue is that unlike supervised training y…

I was surprised by how bullish he is about this. At least a few years ago the experts in the field didn't see AGI anywhere near us for at least a few decades, and all of the bulls were physicists, philosophers or Deepak-Chopra-for-the-TED-crowd bullshit artists who have never written a line of code in their lives, mostly milking that conference and podcast dollar, preaching Skynet-flavored apocalypse or rapture. To s…

As an ML researcher who thinks AGI/HLI is still pretty far away, I don't think there's actually a issue with starting a company that's trying to solve this problem. The thing is that you're going to invent a lot of useful tools along the way. It's going to take you 20-50 years to get HLI, but you're still going to produce a ton of research with a lot of practical uses. The AGI does help build hype and investments too, but the milestones are the important part for the business.

Re: John Carmack's new AGI company, Keen Technologies, has raised a $20M round

#395

Post by version_five »

thewarrior wrote: post #390
godelski wrote: post #371

Earlier quoted context omitted.

> As someone working in AI research Being in the tail end of my PhD, I want to second this sentiment. I'm not even bullish on AGI (more specifically HLI) in 50 years. Scale will only take you so far and we have to move past frequentism. Hell, causality research still isn't that popular but is quite important for intelligence. I think people (especially tech enthusiasts) are getting caught in the gullibility gap.

I know it maybe too much to ask but as a noob to this field I’d love to hear more about why AGI is not feasible even in 50 years.

I've seen a few posts here asking why it isn't possible. Research doesn't work like that, you can't just imagine something is possible and demand reasons why it isn't. How is it possible?

It feels a bit like a lot of these fringe science things (electromagnetic fields causing harm etc) where people's argument is that nobody's shown it isn't true. Well if it's true, propose a mechanism and study it. You're not really doing research if you just make up a conclusion and use the fact that it can't be definitively refuted as evidence for it

Re: John Carmack's new AGI company, Keen Technologies, has raised a $20M round

#396

Post by b20000 »

hansoolo wrote: post #288
b20000 wrote: post #155

I thought AGI meant "adventure game interface". Apparently not, what a disappointment!

I was thinking the same and I am as disappointed as you...

even more confusing because of the name “keen technologies”

Re: John Carmack's new AGI company, Keen Technologies, has raised a $20M round

#398

Post by nine_k »

godelski wrote: post #386
ml_basics wrote: post #248

Earlier quoted context omitted.

I work in AI and would roughly agree with it to first order. For me the key breakthrough has been seeing how large transformers trained with big datasets have shown incredible performance in completely different data modalities (text, image, and probably soon others too). This was absolutely not expected by most researchers 5 years ago.

The problem here is that transformers aren't always the best tool. They are powerful because they can capture large receptive fields but ConvNext even shows that a convolution can still out perform Transformers on some tasks. There's plenty of modern CNNs that have both small and large receptive fields that perform better than transformers. But of course there is no one size fits all architecture and it is fairly nai…

I wonder if combining different approaches works in current AI research. I suppose it should.

A suppose something approaching AGI, an intelligent agent which can act on a _wide_ variety of input channels and synthesize these inputs into a general model of the world, could use channel-specific approaches / stages before the huge "integrating" transformer step.

Re: John Carmack's new AGI company, Keen Technologies, has raised a $20M round

#399

Post by pengaru »

braden-lk wrote: post #329
MichaelCollins wrote: post #283

Earlier quoted context omitted.

From Masters of Doom : Scott Miller wasn’t the only one to go before id began working on Doom. Mitzi would suffer a similar fate. Carmack’s cat had been a thorn in the side of the id employees, beginning with the days of her overflowing litter box back at the lake house. Since then she had grown more irascible, lashing out at passersby and relieving herself freely around his apartment. The final straw came when she p…

I hope he’s maybe grown up since then. Not thrilled to have that attitude from someone working on general AI.

> Not thrilled to have that attitude from someone working on general AI.

Once a technology exists it rapidly changes hands/applications either way. So it makes little/no difference what attitude the creators possess at the time of innovation.

Re: John Carmack's new AGI company, Keen Technologies, has raised a $20M round

#400

Post by Calavar »

Jack000 wrote: post #327
sailingparrot wrote: post #227

Earlier quoted context omitted.

> we didn't have the theory or the models, there was no path forward and so it was impossible to predict. Now we have a fairly concrete idea of what a potential AGI might look like - an RL agent that uses a large transformer. Who is we exactly? As someone working in AI research I know no one that would agree with this statement, so im quite puzzled by that statement.

Which part do you find objectionable - the lack of progress in previous years or the current/future potential of transformers in RL? I do work in ML but mostly applications instead of research.

One big issue is how exactly we'll continue to scale. Exponential growth is hard to maintain.

An example: In the Chinchilla paper [1], the authors suggest that most big transformer models are undertrained, and that we will probably see diminishing returns in scaling up the size of networks if we don't also scale up the size of the datasets. They have a subanalysis where they extrapolate out how big datasets will need to be for larger models. If you believe it, then within two to three generations of big NLG models, we may need on the order of 100 trillion text tokens. But it's not clear if that much text even exists.

[1] https://arxiv.org/abs/2203.15556

Post reply on HN