Does anyone think the current AI approach will hit a dead end? \ stacker news ~AI

pull down to refresh

Does anyone think the current AI approach will hit a dead end?news.ycombinator.com/item?id=45164234

Awesome discussion 🍿

You don’t have to be super smart or waste billions to understand a simple logical fact - you cannot achieve something that isn’t even defined well yet.

The amount of cynicism motivated by amateur hatred is way way less than the amount of optimism generated by professional greed and profit motive.

5 is a major bump-up in terms of version name and yet it is impossible to tell what exactly has improved over the previous version by simply using the product.

0 sats \ 0 replies \ @fourrules 5h

I think it's very good at converting unstructured data to structured data and rather than generating anything new it will help us to navigate a fragmented and noisy internet, which was prohibitively expensive for traditional parse engineering.

Its already extremely cheap to do this, it's the overlooked value proposition because the word "generative" has lead people astray, thinking that AI is going to be a producer of new knowledge and insights.

I think of it like the touch screen, something that had been around for a couple of decades before the iPhone, but took a long time even after the iPhone to really change the way with interfaced with computers.

Right now there is a rush to layer AI on top of products, where as to really shine it needs to be threaded seamlessly into products where appropriate from inception, right through the development process. It is very difficult to retrofit into software systems beyond gimmicks and minor features. I hate the AI bloat as much as anyone.

Yet where it's built into new and novel products it's dramatically opening up the kinds of things developers can achieve.

I am working on solving event aggregation, because event data is fragmented and contained in unstructured sources that are poorly indexed by search engines. Event data is only valid between the announcement and the date of the event, and can change at any time between those dates, making caching tricky. But updates are in WhatsApp and Instagram and email newsletters, so it's all a mess.

LLMs open up this opportunity, $3.2 billion has been burned by startups from Songkick to IRL.com, because before LLMs a feed of event updates has been impossible despite the data being mostly public.

Cheap models with OCR can do it.

There are many many things that LLMs blow right open, it's just going to take time for people like me to build the solutions to previously intractable problems.

So yeah, people are going to be disappointed that vibe coding doesn't make software engineering a trivial pursuit, but then who is to blame for that, AI or people?

0 sats \ 0 replies \ @rootmachine 9h

Yes. It will hit a dead end. It is a matter of months or maybe 1-2 years. The dead end is visible in the way all big names of AI have changed their speech. They will just not have the balls to pin it.

0 sats \ 0 replies \ @jbschirtzinger 11h

In one way or another it will. It's all ready a dead end in that it is making the world of art suck. It is a thieving mimic when it comes to creativity, and that is all it can ever be.