Hacker News Reader

Comment by Synaesthesia | Hacker News Reader

Funny because "e-waste" is also how I would characterise the content that these generative AI models output.

Saying AI will generate tons of e-waste is redundant. AI is waste.

I really don't understand the utter lack of imagination on HN lately. Here's an amazing technology that nobody could even fathom existing five years ago, and people's take is "AI is waste".

Like, really? If you've used ChatGPT and it didn't do anything useful for you, I'm sorry, but the problem was between the keyboard and the chair.

Well, I can understand it. It is an amazing (in the literal sense) technology, and yet after playing with it you come to realise it doesn't really deliver what it promises and is more of a demonstration of the power of enormous amounts of data than it is some kind of breakthrough in artificial intelligence. Also, it's probably mostly a backlash to the constant hype and outrageous claims of impending doom or utopia rampant on Twitter and Reddit; HN seemed to be nothing but AI/LLM talk for pretty much the whole of 2023.

What does ChatGPT do for you?

It writes my code, it explains concepts or context or phenomena to me, it takes notes, it handles my calendar, it generates images, it does boring jobs, along with a very long tail of ad-hoc tasks.

I don't know about the claims, but "the claims are exaggerated" is very different from "AI is waste".

There is also a possibility that the GP just doesn't look/extrapolate into the future farther than a few weeks. That would explain the lack of excitement.

For example, did a month ago they anticipate the computer use tool? When did they expect it to happen?

Given how grievance artists community has been for these AI, it is very easy to understand. The tech is amazing, sure. But I think you are just not seeing the cost.

If I just took on 1 million USD in debt without a clear plan how to repay it. I would not feel optimistic and imaginative about how I can spend this money. I would go into risk averse mindset.

You can ask it about Nietzsche, and it will reply with wonderful and esoteric interpretations of his work, because he is widely studied and ChatGPT has ingested all the works of all Niezsche exegets there are.

You ask it about Hölderlin, and quality drops off, because Hölderlin is not that well studied in the Anglo-American sphere, the majority of Hölderlin studies is in German. It's odd that it cannot translate what it ingested in German and bring it back up in English.

And when you ask it about chemistry ChatGPT is totally lost because it has no idea about concepts, even though the concepts are explained in the textbooks it has read. Until there is a breakthrough that allows AI models to gain understanding all these models are plainly useless.

One can tell - any graduate student in any of the sciences has read way less than ChatGPT but their understanding is infinitely better.

No no, don't get me wrong it's very impressive what "AI" can do. For instance it's ability to look at a huge amount of data and generate useful responses is obviously impressive.

But I do think it's overhyped, and I was referring specifically to the generative AI which although cool, doesn't actually create anything new.

AI might not be waste, but it is wasteful. I mean, at least a few orders of magnitude less efficient than a human. It costs me almost no energy to write an essay, ChatGPT on the other hand...

We should calculate how many calories you burn in the time it takes you to write an essay, versus how many calories ChatGPT burns.

How are you on HN with this kind of attitude? I get that AI is overhyped for a lot of applications but I just shipped an LLM powered feature at $dayjob that saves around 4-8 hrs per use. Our previous version based on NLP had been scrapped for years because it performed so terribly. We just paid humans to do it. The difference between the old and the new is scoring <5% correct to consistently hitting 95%. It's that bird xkcd.

So far, that is the main use case for AI. Which is great, but probably doesn't deserve the trillion dollar hype (and e-waste).

We'll probably get another AI winter once the hype dies down, but we'll have added a few more tools to our toolbelt.

I can agree with you there. I don't think the new use-cases that have been unlocked are world-shattering but they're nonetheless really cool.

We're finally going to have dictation that works reliably, voice assistants that don't suck, really sick photoshop tools, grammar and writing style checkers built into Word that can provide real feedback, a lot of stuff that just wasn't possible before.

Reply on news.ycombinator.com