I agree that AI/ML is overhyped right now. There are however some practical applications for machine learning, even for web browsers. The most obvious one to me is translating content in foreign languages. Another would be tools that make poorly designed websites more accessible for the visually impaired.
As someone who is fluently bilingual I have tested the exact thing people say AI in the browser would be good for, translations and subtitle generation and it was abysmal. It produced sentences that absolutely look correct but the meaning has fundamentally changed. Someone fluent in both languages would know what it intended to say but for those people we wouldn't need the translation anyway.
It was deemed far too much work to be viable at this point seening as we would still need to employ someone to double check the translations anyway so the project was canned.
Yeah, it’s not really enable new projects in the language arts unless it’s cases where accuracy and the truth don’t matter compared to volume (and those are mostly things that are harmful to society).
You can now write a non-fiction book with a subject matter expert and an editor, skipping the ghost writer. Saves one whole salary, right? No, because it’ll take a lot more time from both the SME and editor to reach the same quality level as before. You can dial the quality down a little—worse presentation of material, more repetition, more inaccuracies—and save some money, which was harder to reliably do before, so there’s that I guess, but now we’re back at “it saves money if you’re trying to make junk”.
I don't know how you can say that. We passed the Turing test within the past year, and nobody even noticed because that now seems like such a low bar for where we'll be in a few years.
ML is an important component of AI, but it's not a simple "rebrand". I get that it's easy to think of it as marketing hype, because there's a lot of people (the same people who got on the blockchain bandwagon) shilling crap. But what we're seeing isn't overhyped; it's severely underhyped.
> I don't know how you can say that. We passed the Turing test within the past year
Easy: Remember OpenAIs claims regarding the bar exam and did you read the news the last couple of days how it was all bollocks?
Great... it's another hype. It's "just" statistics, very impressive, but as soon as it will run out of training data, it will plateau... and it will run out by GPT-5 or GPT-6 :)
Then it can learn from its predecessor's hallucinations maybe :D
> But what we're seeing isn't overhyped; it's severely underhyped.
I don't know if passing the turing test is really that much of an accomplishment. Some would argue we create software that passed the turing test when Eliza was made, which was developed in the 60s. I mean, Eliza really kind of shows that the turing test isn't actually much of a test. Your testing more of how easy it is to fool a person and exactly how intelligent or smart an artificial system is.
Agreed, which is why the next part of my sentence went on to say "nobody even noticed because that now seems like such a low bar for where we'll be in a few years"