AGI Is Being Achieved Incrementally | Hacker News Reader

AGI Is Being Achieved Incrementally

latent.space

21 pointsswyx14 comments

I think it's less that artifical general intelligence is being achieved and more that artifical intelligence of a different color is being tripped over.

Given that a lot of HN users don't read a 5 paragraph article before commenting on it, I wonder how many will listen to a 2+ hour podcast. I would like to hear from someone who has listened to the whole thing, but maybe there should be a post called "Ask HN: Is AGI being achieved incrementally?"

Betteridge's Law of Headlines will answer your reformulated post title accurately: no.

The law doesn't apply:

> Any headline that ends in a question mark can be answered by the word "no".

It's a statement not a question.

The reformulated headline in the comment is a question, so that does apply to the reformulated one... which the quote was in response to.

Well, actually, the reformulated one was for an "Ask HN" submission title, not a headline. ;)

> Headline: a heading at the top of an article or page in a newspaper or magazine.

I don't think an "Ask HN" fits the definition.

I'd like to point out that Betteridge's Law is, like any other such "law", kind of tongue-in-cheek. It's not serious.

The real point of my earlier comment was that I was voicing an opinion that it is not the case that we are "achieving AGI incrementally", and I referenced Betteridge's Law as a way of making that point.

This is just an intuition on my part. AI will make steady progress, but AGI will not be realized til we better understand consciousness. There will be progress in domain specific AI, but general intelligent autonomous action requires that the lights be on, that is, consciousness.

Huh, I think the exact opposite. We don't understand consciousness, so I expect it to be something accidental that ends up creating it. A misconfiguration possibly, or just a couple of tools being hooked up in a novel way. We probably won't know it's conscious for the first couple years.

Kinda weird to create consciousness when you don't understand what it is, but neither did the universe and we still ended up conscious.

This isn't possible without being able to identify an AGI, which we are completely unable to do. That's one of the great open questions, and we've spent ~3,000 years on it.

Without a definition, you can't say if we're remotely close.

Saying we spent 3000 years on it a stretch. The modern concept of intelligence is itself barely two hundred years old [1]. Trying to see how it might apply to a machine is really just a bit of poetic exercise and language development.

We’re already at a point where some people will start staunchly attributing AGI to extant technology and others will disagree, and the tides will shift in favor of the former as technology further develops and the (themselves very modern) traditionalists tire of fighting about it.

You’re thinking about the question scientifically, but neither language or culture work that way.

There is very likely to be a broad but loudly disputed consensus that “AGI has arrived” within the next few decades, regardless of what more the technology can actually do.

Academics will be writing about it as such, corporations will be asserting as such, legislatures will be codifying it as such, budding pseudo-religious movements will be vaunting it as such, and the many old cynics like you and me will just be saying “WTF?!” before resigning ourselves to watching Seinfeld reruns and feeding birds in our own private peace.

[1] Danziger, K. (1997). Naming the mind: How psychology found its language.

So what definition of AGI (N.B. not superintelligence) excludes GPT-4? It's good at a wide range of tasks without being specifically trained on those tasks. It's pretty dumb at some tasks, but aren't we all?

Serious question. I think we've invented AGI, but that term is now useless and we need to come up with better terms such as superintelligence, or maybe a numeric scale.

The podcast episode doesn't seem to have anything to do with the title unless I'm missing something?

They don't seem to discuss AGI, just a general discussion of OpenAI DevDay.

… what remains to be seen is how many human lifetimes those increments will consume, and how many cul de sacs we’ll discover along the way.

Reply on news.ycombinator.com