Hacker News Reader

Comment by josefritzishere | Hacker News Reader

Frankly, most AI is pretty meh, really only impressing AI enthusiasts, with their near religious fervor. There's a big delta there between power users and everyone else. It writes bad emails, messy code, incorrect metrics, dubious slide decks, ugly graphics... the consumer rejection of AI is only growing. At the same time, the product itself is sold wildly below cost. So it needs to be increase in price at the same time users are rejecting it. This is only ending with a huge market correction.

View on HN

How do you know it's being served at wildly below cost? AFAIK API pricing across the board is actually higher than what inference would cost? Yes there are subscription plans that subsidise tokens but they exist primarily to hook people in, so as to encourage their workplace to adopt it, whereby they then have to pay for an enterprise plan with zero subsidies like the subscription plans have. Training is very expensive, but you only do that once.

I predict costs will fall.

That’s roughly it.

I don’t really have a use for it myself that I can’t solve some other way cheaper or easier without the damage and uncertainty it does.

I’m not really that impressed.

It seems by definition "below par" of whatever human activity it's training on. How can we achieve greater and greater heights of intelligence if we're just creeping closer and closer to a sub-par copy of human creative activity.

It seems by definition "below par" of whatever human activity it's training on

How many sub-par mathematicians could solve the unit distance conjecture?

I never understood people like you.

Factually that's just not true. Five years ago current llms would have seen like magic to even data scientists working on them and realistically their impact has been felt by many. Just taking jobs into account, imagine an average worker in an average industry; how do you think something like Fable compares? Worse? By how much?

Like even many years ago we had alphafold that had a non negligible impact on biology, who knows what's gonna happen when this technology grows up.

Sorry but it's painfully true. I now spending time daily like the "reverse-centaur" fixing things that AI has hallucinated, conflated, or just done wrong. It's miserable work. The copy/pasta part is easy.

My point is you'd be doing the same if you wrote the code or an average person did it. If you spend time for the prep work AI is solid in execution.

I am at this very moment re-writing AI slop that contains errors a half-competent intern would have caught. It's demoralizing that this is the future.

Weird thing I've encountered. I've seen all the stuff you say, but not all at the same time.

The stuff I'm getting from ChatGPT this week has been absolutely garbage. Back in the very early days you might prompt 5 times and see if it was consistent with itself in the conclusion, even if not how it got there; this week it's not even been that. But back in those days the UI was also fast, today it lags massively on relatively short chat sessions.

Ugly, dubious, incorrect, it definitely feels like a regression. And its love of emoji hasn't gone away.

Claude no longer works for me on Safari, only Chrome. No lag at least, but the free message allowance has shrunk a lot.

I'm not too fussed about messy code, given the humans I've worked with, but that's about it.

Even the ChatGPT image model is… for all the improvements in the model itself, the UX isn't good enough to make up for what the model still can't do. In many cases I actively prefer running Stable Diffusion locally because that's easier to fix the last 5% than having to deal with a completely different 5% wrong each time.

But yeah, correction very possible. Was thinking so just on the basis of the % of US electrical power being called for: that can't possibly be sustainable for the broader economy.

I'm not an AI advocate by any means, but its usefulness is not all or nothing. I use it at work a lot - scanning our documentation + data warehouse + version control system all at once saves me a lot of time. I can let it build boiler plate code (via skills) and the actual code implementation as a template/POC for me to then edit, again saving a lot of time.

Outside of my work - AI video has reached a point where creative artists are generating engaging, realistic content that is genuinely entertaining. IMO the common thread is that in the right hands it can produce amazing things. For anyone without the expertise, it defaults to slop.

Reply on news.ycombinator.com