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

AGI Overhyped?

news.ycombinator.com

Post reply on HN
Print view View original

21–30 of 36 posts

Re: AGI Overhyped?

#21

Post by tylerhou »

jstanley wrote: post #14
tylerhou wrote: post #9

> AI already blows most people out of an IQ test at a fraction of the computational power of a brain AFAIK, IQ tests used in psychological evaluations do not contain any randomness so exact answers are almost always in distribution. I haven't seen someone compare AI to an IQ test that is not in distribution. On ARC-AGI, which is mildly similar to a randomly generated IQ test, humans still are much better than LLMs. h…

I scrolled down but didn't find a chart comparing average human performance to AI performance. The only chart I found was comparing the costs of different models.

Sorry, you're right that the chart on the home page does not have human performance. The leaderboard chart does: https://arcprize.org/leaderboard. And the leaderboard by default shows scores for ARC-AGI 1 and 2. The models are much worse at 2 than 1; the best performing model scores around 15% (Grok 4, thinking), while humans are at ~100%.

Re: AGI Overhyped?

#23

Post by dudefeliciano »

mg wrote: post #18
ngruhn wrote: post #5

> I've never heard a decent definition I would say AGI can do every intellectual task a human can do. Maybe the raw cognitive power is already there but scale matters. If it can't work on a task for months on end, it's not AGI. Because humans can do it. > AI already blows most people out of an IQ test at a fraction of the computational power of a brain Does it? I thought the brain is much more energy efficient.

> it can't work on a task for months reply = AI.ask(task) while true: reply= AI.ask(f""" Improve the reply below: Task: {task} Reply: {reply} """") print(reply)

have you tried this? honestly curious at what kind of hallucinations would come out of this

Re: AGI Overhyped?

#24

Post by SCdF »

mitthrowaway2 wrote: post #16
SCdF wrote: post #8

Part of the difficulty here is that new age AI / LLMs in discussion has a lot of similarities with crypto, in the sense that there is a lot of nonsense out there. Unlike crypto there is obviously some value, as opposed to none, so it's not all grift. But like you I find it hard to sort the difference between the two. Fundamentally for me I can't get over the idea that extraordinary claims require extraordinary eviden…

I'm not sure. We never expected that there even could be a magic trick that came so close to mimicing human intelligence without actually being it. I think there's only so many ways that matter can be arranged to perform such tricks, we're putting lots of work into exploring that design space more thoroughly than ever before, and sooner or later we'll stumble on the same magic tricks that we humans are running under…

> and sooner or later we'll stumble on the same magic tricks that we humans are running under the hood.

Right, so this is the extraordinary claims bit. I'm not an expert in any of the required fields, to be clear, but I'm just not seeing the path, and no one as yet has written a clear and concise explainer on it.

So my presumption, given past experience, is that it is hype designed to drive investment plus hopium, not something that is based on any actual reasoned thought process.

Re: AGI Overhyped?

#25

Post by ben_w »

SCdF wrote: post #8

Part of the difficulty here is that new age AI / LLMs in discussion has a lot of similarities with crypto, in the sense that there is a lot of nonsense out there. Unlike crypto there is obviously some value, as opposed to none, so it's not all grift. But like you I find it hard to sort the difference between the two. Fundamentally for me I can't get over the idea that extraordinary claims require extraordinary eviden…

> It's like saying that if a magician worked _really really hard_ on improving, evolving and revolutionising their "cut the assistant in half and put them back together" trick, they'll eventually be able to actually cut them in half and put them back together.

So, surgery?

As the stage magicians Penn and Teller (well, just Penn) said, stage magic is about making something very hard look easy, so much so that your audience simply doesn't even imagine the real effort you put into it.

Better analogy here would be asking if we're trying to summon cargo gods with hand-carved wooden "radios".

Re: AGI Overhyped?

#26

Post by ben_w »

SCdF wrote: post #24
mitthrowaway2 wrote: post #16

Earlier quoted context omitted.

I'm not sure. We never expected that there even could be a magic trick that came so close to mimicing human intelligence without actually being it. I think there's only so many ways that matter can be arranged to perform such tricks, we're putting lots of work into exploring that design space more thoroughly than ever before, and sooner or later we'll stumble on the same magic tricks that we humans are running under…

> and sooner or later we'll stumble on the same magic tricks that we humans are running under the hood. Right, so this is the extraordinary claims bit. I'm not an expert in any of the required fields, to be clear, but I'm just not seeing the path, and no one as yet has written a clear and concise explainer on it. So my presumption, given past experience, is that it is hype designed to drive investment plus hopium, no…

Sure, but evolution isn't an actual reasoned thought process, and it still managed us without even having humans as an explicit goal, we just popped out of the process by accident as a way to be effective at surviving in the wild.

Re: AGI Overhyped?

#27

Post by trabant00 »

I define AGI as the only real AI that can exist. Anything less is just mimicry. To give a stupid/simplified example: a truck driving specialized "AI" would not be able to decide when to stop or not when somebody steps in front of it, making it trivial to rob "AI" driven trucks. To decide it needs to understand the kinds of people that exist, their motivations, the laws, etc. So it needs to be an AGI. Otherwise it will make horrible mistakes we don't even think are possible, or are so uncommon that when they happen to a human they make the news.

Re: AGI Overhyped?

#28

Post by jstanley »

tylerhou wrote: post #21
jstanley wrote: post #14

Earlier quoted context omitted.

I scrolled down but didn't find a chart comparing average human performance to AI performance. The only chart I found was comparing the costs of different models.

Sorry, you're right that the chart on the home page does not have human performance. The leaderboard chart does: https://arcprize.org/leaderboard . And the leaderboard by default shows scores for ARC-AGI 1 and 2. The models are much worse at 2 than 1; the best performing model scores around 15% (Grok 4, thinking), while humans are at ~100%.

Thanks, and do we know if the humans are average people off the street, or unusually-intelligent people?

EDIT: OK, I see there are 3 types of humans:

"Avg. Mturker" does worst. "Stem Grad" and "Human Panel" are basically equivalent in terms of quality but differ in cost.

It's not obvious to me whether an average Mturker would be more or less clever than the average person. Mturk doesn't pay very well, so you'd think you'd have to be below average to want to do it. But potentially it attracts people of above-average intelligence who just happen to live in the third world?

Re: AGI Overhyped?

#29

Post by rsynnott »

jstanley wrote: post #28
tylerhou wrote: post #21

Earlier quoted context omitted.

Sorry, you're right that the chart on the home page does not have human performance. The leaderboard chart does: https://arcprize.org/leaderboard . And the leaderboard by default shows scores for ARC-AGI 1 and 2. The models are much worse at 2 than 1; the best performing model scores around 15% (Grok 4, thinking), while humans are at ~100%.

Thanks, and do we know if the humans are average people off the street, or unusually-intelligent people? EDIT: OK, I see there are 3 types of humans: "Avg. Mturker" does worst. "Stem Grad" and "Human Panel" are basically equivalent in terms of quality but differ in cost. It's not obvious to me whether an average Mturker would be more or less clever than the average person. Mturk doesn't pay very well, so you'd think…

Additional caveat: some of the "avg mturker" cohort are almost certainly using LLMs to participate.

Re: AGI Overhyped?

#30

Post by SCdF »

ben_w wrote: post #25
SCdF wrote: post #8

Part of the difficulty here is that new age AI / LLMs in discussion has a lot of similarities with crypto, in the sense that there is a lot of nonsense out there. Unlike crypto there is obviously some value, as opposed to none, so it's not all grift. But like you I find it hard to sort the difference between the two. Fundamentally for me I can't get over the idea that extraordinary claims require extraordinary eviden…

> It's like saying that if a magician worked _really really hard_ on improving, evolving and revolutionising their "cut the assistant in half and put them back together" trick, they'll eventually be able to actually cut them in half and put them back together. So, surgery? As the stage magicians Penn and Teller (well, just Penn) said, stage magic is about making something very hard look easy, so much so that your aud…

No, not surgery. Surgery wasn't gotten to by way of working really hard on a magic trick. I'm also reasonably sure surgery is not at a point where you can cut someone entirely in half and put them back together.

Since you brought up Penn and Teller, take the bullet catch. They are not actually catching a bullet. No matter how hard they work on perfecting their bullet catch trick, this will not allow them to catch a real bullet shot from a real gun in their real teeth. Working on the trick is not the journey where the end point is doing the actual thing you're representing in the trick.

Post reply on HN