@codinghorror I think inference will get much cheaper. Companies haven't really had much of an incentive to make more efficient models because they were funded completely on investor hype. They all were competing with each other on who could produce the best model no matter the cost (framed as the race to AGI).

IMO, even the small 27B and 35B Qwen models are decent if put into a good harness (e.g. Opencode), and they run fine on a 24GB consumer GPU.

I think I've read that the AI bubble is roughly 2x-3x the size of the dot-com bubble, so don't know if it will be less traumatic.

@codinghorror if, as you say, these models are so good, why haven't I seen a single good feature or update yet?

On the other hand, the products I know and love are going to dogs.

I have a support ticket in AWS, open as we speak. 7 days since I got any reply from them. The automated AI generated response was no good. There seems to be no human fallback so the ticket is still unassigned.

I keep asking people and I get the same story everywhere across all tech products.

@codinghorror @briankrebs the trauma will come when claud stops working and all these companies are stuck with codebases that aren't human maintainable

@codinghorror There is a lot of wisdom in what you say. That's the worst part of the AI overhype and the scammers who are pushing it. It is making me avoid a technology that has some usefulness because they make it impossible to get that utility without their bullshit.

@codinghorror I personally expect it to be MORE traumatic, because of the staggering amount of money involved, the relative size and position of key players, and the level of exposure how many people have to that financial surface.

@codinghorror Looks like Tulip Mania to me.

@codinghorror The thing about the dotcom boom is that, while many companies were overvalued, it was based on relatively affordable technology with a proven revenue stream. Even back then, buying a domain name and a server ultimately cost hundreds, not billions. While there were certainly a lot of scammy online businesses, most of the sites the average person interacted with every day had an audience and had revenue that could cover their core operating costs. Plenty that didn't have that went bankrupt, and the rest had to massively curtail their growth, but underneath that investment bubble were a lot of fairly sound businesses.

But this "AI" boom? The models that the average person interacts with every day are losing billions of dollars a year. When the investment bubble pops, all of those companies are dead. Some form of the technology will absolutely continue, doing specialized analysis just as it existed before the bubble, but I don't see how the more ubiquitous forms can possibly survive.

When the dotcoms popped, the average person just kept using dotcoms. We barely noticed the change. But I doubt the average person will still be using "AI" after this one fails.

@codinghorror will those existing use cases remain if the cost to use these models increases 10x to reflect their actual cost? In terms of compute cost, LLMs aren't following the usual Moores Law curve, capability has pretty much scaled with cost... we've just found faster ways to burn tokens.

@codinghorror @briankrebs isn't the issue that AI as it is now isn't profitable? The services we are getting used to are merely promotional and in truth it should (and will) be a lot more expensive in the future? This is the drug dealer's free hit, the time where we can consume without consequences is fast approaching an end.