Comment by chomp | Hacker News Reader

Comment by chomp | Hacker News Reader

Parent Full thread

I wanted to write a comment like this but feared it would be too negative or dismissive. But I agree with the general thrust of your argument. The article is a little outdated because a lot of the examples of deficiencies in the article, work in GPT4 (I tested a couple). But maybe the author is sandbagging ChatGPT to bring the non-tech people back to Earth.

When ChatGPT was released, a lot of people (even on this site) were trying to get it to think abstractly and were applying human qualities to the bot, when it's relatively clear that it's all just fancy statistics. Even today, when talking to non-tech people, there's still a lot of people who believe it's a magical thinker-in-a-box. Someone has to talk those people down, and it ain't gonna be me.

It's not perfect, but it doesn't have to be to have applications IRL. Yeah, the web chat is meh whatever, but the API is where it actually shines. It's the integration with existing business processes and code that allow new, more human ways of interfacing with systems. Asking it for what it knows in its statistical blob, probably will lead to heartache. Giving it some context and asking it "what can you infer from this?" is where it's strongest.

> when it's relatively clear that it's all just fancy statistics. So just like human brain then.

> So just like human brain then.

Do you have evidence to substantiate that claim, or are we supposed to just take it on faith?

Quantum mechanics is “just” fancy statistics. Unless we believe in some non-physical/supernatural phenomenon going on in the human brain, this sort of reductive argument works as well for the human brain as it does for LLMs (in other words, not very well. Sufficiently complicated statistics is what underlines everything, as far as we can tell.)

I actually agree that I don't think there's anything particularly special about our meat and chemical GPUs, but analog's fidelity has historically been difficult to replicate precisely in digital space. Maybe since we don't care for emotion in our intelligence, we don't need the analog. But I think our brains are a little more (but not that much more) complex than fancy statistics. After all, the math models the real world, not the other way around.

I wrote a massive rant for a response but then I deleted it.

I feel sorry for people think about themselves this way, why just compare yourself to a calculator?

If it was possible to build an artificial brain that was self-ware, it would probably want to be a lot more like us than you think. To feel and connect, and experience love and what that’s like is a very special part of existence.

There seem to be a lot of people in this thread who consider themselves to be calculators. I too have to restrain myself from writing massive rants.

People espousing this opinion seem to have grown quite a lot since ChatGPT first became public. Cynical as I am, I do suspect some astroturfing trying to prepare people for being "obsolete".

I also suspect astroturfing, particularly here, and I also notice it on Substack (I don't visit a whole lot of other sites anymore). On Substack I see two clusters, one group _very into_ AI, the other advocating a lot of caution.

I don't think it's preparing people to be obsolete; I think it's a combination of AI startup founders hyping up their tech, and people who identify with big tech/"the side that wants AI".

Hype no doubt plays a part, which is a perfect cover for any truly malicious astroturfing. But when I say "prepare to be obsolete", I mean prepare to be obsoleteD.

What’s the difference ?

Right now, billionaires who profess the view that AI could end humanity are working their hardest to progress the field as fast as possible, scoffing at "ethics", "safety", and "responsibility". So if they do get their god-machine, they're not going to look around at 8 billion fellow humans (biological stochastic parrots, according to some) who are suddenly "obsolete", but using up resources, and think "fully automated space luxury communism".

Got it. Definitely feels like some type of play for dominance.

Which leads to the questions, dominance for whom, and over whom. And why, if we're supposedly entering an age of post-scarcity where AI can enable anyone to do anything, we would need dominance in the first place.

I mean, I think distilling my argument down into "I'm not much more than a calculator" is a little bit much, but maybe I was unclear. I just don't think there's magic there. Brains are mysterious, but so are our LLMs, and we invented those. Most of what we experience emotionally isn't magic, it's a complex chemical reaction of hormones applied to our inputs and outputs. But like I said, I don't know if we'd value that in our AI or not, so that complexity may not be necessary to emulate.

I’m going to go out on a limb here but I bet that deep down inside you know that intellect is only one mechanism from trying to understand the infinite complexity of existence.

Reply on news.ycombinator.com