John Carmack's AGI Startup Keen Raises $20M From Nat Friedman, Daniel Gross and Sequoia 44
John Carmack, a programmer who founded gaming firm id Software and served as chief technology officer of Oculus, has launched a new artificial general intelligence startup called Keen Technologies, and it has raised $20 million in a financing round co-led by former GitHub chief executive Nat Friedman and Cue founder Daniel Gross, Carmack said Friday. Stripe co-founder Patrick Collison, Shopify co-founder Tobi Lutke, storied venture fund Sequoia and microprocessor engineer Jim Keller also invested in the round, a name of which as well as the startup's valuation Carmack did not disclose. In a Twitter thread, Carmack adds: This is explicitly a focusing effort for me. I could write a $20M check myself, but knowing that other people's money is on the line engenders a greater sense of discipline and determination. I had talked about that as a possibility for a while, and I am glad Nat pushed me on it. I am continuing as a consultant with Meta on VR matters, devoting about 20% of my time there.
Re: (Score:3, Funny)
Well, in the worst case it can be smarter than you, which might barely be good enough.
Re: (Score:2)
So far, once you define a specific task.. such as handwriting recognition, or even image recognition. Computers seem to be better than human. At some point every task will be put in the "specific task" bucket. It will take time though.
Re: (Score:2)
So far, once you define a specific task.. such as handwriting recognition, or even image recognition. Computers seem to be better than human.
Really? Because I've not experienced that with my own handwriting. Even something simpler like OCR often falls short. I've been using OCR a lot and I find that it still fails to recognize some very obvious text even at 1200 DPI. I know I don't have the cutting-edge technology, but I think it's premature to declare that computers have surpassed human competence here.
Re: (Score:3)
Amazon's can read handwriting that *I* can't.
Re: (Score:2)
Actually OCR is still not very good. It fakes you out by neglecting the plausibility test and sometimes delivering complete bullshit with confidence.
Re: (Score:2)
And no, it doesn't deliver complete bullshit. We are a commercial client, and like I said, it outperforms humans without even factoring in the time differential.
I can only surmise that you haven't used a modern ML OCR.
Re: (Score:2)
So far, once you define a specific task.. such as handwriting recognition, or even image recognition. Computers seem to be better than human.
Really? Because I've not experienced that with my own handwriting. Even something simpler like OCR often falls short. I've been using OCR a lot and I find that it still fails to recognize some very obvious text even at 1200 DPI. I know I don't have the cutting-edge technology, but I think it's premature to declare that computers have surpassed human competence here.
Computers have not. For anything that requires correction by insight, computers are a complete failure so far, specifically because of the absence of AGI and the extreme limits of AI. Humans do it all the time automatically. Machines cannot do it. And the optical analysis alone is not enough in many cases or takes to long if you do all that is possible.
Re: (Score:2)
Some of them, yes. But that is automation (often misnamed "AI") and very much _not_ AGI.
Re: (Score:2)
At some point every task will be put in the "specific task" bucket.
That's not what GAI is about. A GAI should be competent at finding and filling its own buckets. That's what the 'general' part is all about. It should understand new problems and formulate a solution.
Re: (Score:2)
Yes, that is pretty much the point. If it is a known field and a known taks, basically a "book" (i.e. a really large table) can do it. AGI is when you go beyond the need for the field to be understood and the rules to be known and the machine starts to get actual insights. So far only humans (well, some humans and probably only 15% or so) can do it.
Re: (Score:2)
Re: (Score:2)
At this time, nobody knows whether AGI is even possible in this universe. This is either obviously fraudulent to the proponents are deep in delusion.
Yes, that's why he went with the most naive "investors" he could find.
Re: (Score:2)
At this time, nobody knows whether AGI is even possible in this universe. This is either obviously fraudulent to the proponents are deep in delusion.
Yes, that's why he went with the most naive "investors" he could find.
Pretty much, yes.
Re: Obvious fraud (Score:2)
Re: (Score:2)
This is only half true. AGI is impossible with current iseas, but not overall impossible.
That is completer bullshit. At this time it is completely unknown whether AGI is possible and any claim that it is represents a statement of belief with no scientific validity.
Re: (Score:2)
At this time it is completely unknown whether AGI is possible
Well, we humans posses a certain amount of general intelligence. It seems that reality tells us that it is possible.
Re: (Score:2)
Only if you subscribe to the pysicalist faith, which is religion, not Science. Science does not try to do invalid conclusions by elimination. Science only allows conclusions by elimination if _everything_ about a subject is known. That is very much not the case for physical reality.
Re: (Score:2)
Science does not try to do invalid conclusions by elimination. Science only allows conclusions by elimination if _everything_ about a subject is known. That is very much not the case for physical reality.
With such argumentation you're really killing everything that can ever be accomplished by science.
You could assume that there is a world around you (which is the basic assumption for science to be valid) but you can't ever be sure. Is that uncertainty a good reason to stop researching this world? Does that make science 'obvious fraud'?
Re: (Score:2)
You do realize you're talking about a man who has demonstratively and repeatedly overcome seemingly insurmountable hardware limitations with amazing application of brilliance?
Those discrete GPU's we have today, which are now powering AI, mining, etc.--you do know who's software prompted Voodoo to make some of the first GPU's, don't you?
Re: (Score:2)
Carmac has done some impressive things in performance. He has not contributed anything fundamental.
Re: (Score:2)
At this time, nobody knows whether AGI is even possible in this universe.
True. But we also don't know that God exists. Proceeding as if he doesn't is still wise, for the very same reason: There's not one single shred of evidence to indicate that AGI isn't possible.
This is either obviously fraudulent to the proponents are deep in delusion.
Now this is strange. You go from saying "nobody knows if it's possible..." to claiming that those who think it might be are delusional.
If I had to guess, you've got a very strongly held belief that AGI isn't possible, and you're bristling at the idea of someone like Carmack thinking it is.
Re: (Score:2)
At this time, nobody knows whether AGI is even possible in this universe.
True. But we also don't know that God exists. Proceeding as if he doesn't is still wise, for the very same reason: There's not one single shred of evidence to indicate that AGI isn't possible.
And fail. The argument works in one direction (assuming non-existence in the absence of evidence after a reasonable intense search). It does not work in the other (assuming something is possible in the absence of evidence after a reasonably intense search). It is a good idea to keep looking, but expecting to be successful is foolish. In both cases. While the search for "God" has gone on much longer with zero results, the search for AGI has also gone on for quite a while (>50 years) and also had zero resu
Re: (Score:2)
the search for AGI has also gone on for quite a while (>50 years) and also had zero results.
You say that, but that search is pretty much limited by our technology and by our ability to analyze intelligences. The recent progress in machine learning comes down to advances in GPU technology and i expect this type of processing to be just a small part of an actual GAI.
What i'm trying to say is that the search we have managed so far is far from exhaustive. It may have taken a long time but that says nothing about what could be uncovered in the future. We're basically still at the baby step phase.
The bi
Re: (Score:2)
I am not "saying that". I am stating that scientifically, there currently is zero results. Do you have any actual results that you can point to?
You also seem to be under the misapprehension that the "AI" field could deliver AGI. It cannot and it is not actually trying to. At least the competent researchers in this area have stopped trying to do that a few decades back. "AI" (which would still be named "automation" if the marketing-assholes had stayed out of it) is known and understood to not be able to do A
Re: (Score:2)
I am stating that scientifically, there currently is zero results.
I agree with this. But that ignores the fact that we do have an example of general(ish) intelligence, which is us.
This distinguishes the whole pursuit of GAI from the pursuit of god on a fundamental level.
You also seem to be under the misapprehension that the "AI" field could deliver AGI.
Well, frankly i think it will necessarily be a cross-disciplinary breakthrough, but i do see current machine learning research as elemental for future discoveries.
is known and understood to not be able to do AGI these days.
While true, this kinds of machine learning is used by researchers who do seek AGI.
You also seem to think that it is a problem of missing computing power. It is not. Otherwise we would already have very, very "slowly thinking" AGI that was pretty dumb, but AGI nonetheless.
That would only be true if it was a solved field. It is far fr
Re: (Score:2)
this is fraud or delusional
Why would Carmack set up a scam?
Re: (Score:2)
this is fraud or delusional
Why would Carmack set up a scam?
I have no idea. For him I would suspect delusional.
Re: (Score:2)
And fail. The argument works in one direction (assuming non-existence in the absence of evidence after a reasonable intense search). It does not work in the other (assuming something is possible in the absence of evidence after a reasonably intense search). It is a good idea to keep looking, but expecting to be successful is foolish. In both cases. While the search for "God" has gone on much longer with zero results, the search for AGI has also gone on for quite a while (>50 years) and also had zero results.
The only failure here is the inference engine you call your brain.
What you have stated is in fact my precise justification for an attempt.
No, it is not. They are doing a start-up. A start-up is a short-term bet on something working in the near future, typically less than 10 years. As even a lab demo is missing, this is fraud or delusional. Delusional would be if the actors here think that AGI has already been demonstrated. Context is important.
Yes, it is strange. Because you moved your own goalposts in your argument.
Trying to fix that by redefining what start-up means is laughable.
A startup is not a short-term bet on something working in the near future.
A startup is any kind of business in its early stages looking for outside funding with a high risk of failure.
Sounds to me like this fits the bill pretty
AGI not explained, names of rich people is ? (Score:2, Insightful)
Re: (Score:1)
Re: (Score:2)
Re:AGI not explained, names of rich people is ? (Score:4, Insightful)
There was a time that Carmack was almost worshipped on Slashdot. Have things really changed to such an extent that he is now derided here as "unimportant rich people"?
Re: (Score:2)
It's literally in the *first sentence* of TFS...
Re: (Score:1)
A real editor would know that.
Try NGI first (Score:3)
I'm still waiting for humans to achieve natural general intelligence...
knowing that other people's money is on the line (Score:2)
John Carmack...Keen... (Score:2)
Will there be another Commander Keen game or not?
A bit of an undersell (Score:2)
Describing him as "a programmer who founded id and was CTO of Oculus" kinda skips over all the contributions he's made to two fairly enormous fields, regardless of how much (or how little) of that will transfer to the AGI space.
We're still basically no closer to AGI than we were 50 years ago, though at least ML has advanced massively in the last decade. It won't hurt to have someone who's not just exceptionally smart but also has a track record of actually delivering get involved. If nothing else, we know t