Hacker News Reader

Comment by lowsong | Hacker News Reader

> the future of software engineering will inevitably have more AI within it

Probably not. We're deep in the hype bubble, so AI is strongly overused. Once the bubble pops and things calm down, some use-cases may well emerge from the ashes but it'll be nowhere near as overused as it is now.

> AI has become a race between countries and companies, mostly due to status. The company that creates an AGI first will win and get the most status.

There's a built-in assumption here that AGI is not only possible but inevitable. We have absolutely no evidence that's the case, and the only people saying we're even remotely close are tech CEOs who's entire business model depends on people believing that AGI is around the corner.

View on HN

> We're deep in the hype bubble, so AI is strongly overused

I don't think these things are really that correlated. In fact, kind of the opposite. Hype is all talk, not actual usage.

I think this will turn out more like the internet itself. Wildly overhyped and underused when the dotcom bubble burst. But over the coming years and decades it grew steadily and healthily until it was everywhere.

Agreed re: AGI though.

That is not how the dotcom bubble burst. Internet usage was growing fast before, during, and after the bubble. The bubble was about silly investments into it that had no business model - that investment insanity is what burst, not overall usage.

I think I am saying the same thing. My comment about "underused" should have been "underused relative to the investment dollars pouring in"

Many of the business models were good too but they had the timing wrong.

Petfoods.com IPO for about $300 million. $573 million adjusted for inflation.

Chewy is at a 14 billion market cap right now.

I think comparing LLMs and the dotcom bubble is just incredibly lazy and useless thinking. If anything , all previous bubbles show is what is not going to happen again.

> I think comparing LLMs and the dotcom bubble is just incredibly lazy and useless thinking. If anything , all previous bubbles show is what is not going to happen again.

Curious to hear more here. What is lazy about it? My general hypothesis is that ~95% of AI companies are overvalued by an order of magnitude or more and will end up with huge investor losses. But longer term (10+ years) many will end up being correctly valued at an order of magnitude above today's levels. This aligns perfectly with your pets.com/Chewy example.

Certain parts of what we call AI will definitely be used more in the future: facial recognition, surrogate models, video generation.

I don't, however, see LLMs as consumer products being that prevalent in the future as currently. The cost of using LLMs is kept artificially low for consumers at the moment. That is bound to hit a wall eventually, at the very least when the bubble pops. At least that seems like an obvious analysis to make at this point in time.

The cost angle is interesting, I'm not close enough to the industry to say for sure. But from what I'm reading inference and tokens are only getting cheaper.

Regarding usage - I don't think LLMs are going away. I think LLMs are going to be what finally topples Google search. Even my less technical friends and acquaintances are frequently reaching for ChatGPT for things they would have Googled in the past.

I also think we'll see the equivalent of Google Adwords start to pop up in free consumer LLMs.

> The cost of using LLMs is kept artificially low for consumers at the moment.

If the results of current LLM performance is acceptable, costs to achieve these same results will inevitably go down as semiconductor process improvements bring markedly reduced operational expenses (power, density, etc.)

Reply on news.ycombinator.com