"AI" is unprofitable and almost completely useless. With capex demands ever increasing just to push the diminishing scaling returns further, this is not sustainable. Their revenue will collapse in the upcoming recession and you'll see it's PE balloon into the hundreds.

> "AI" is unprofitable and almost completely useless.

Famous people said that about the Internet as well, back in the day. And about AMZ.

I've still never bought a single physical product from Amazon.

Only AWS, which I stopped using when they decided to raise their price I'd already locked in and prepaid for, which should be unlawful. Then I realized every traditional provider is much cheaper.

> I've still never bought a single physical product from Amazon.

In the context of this thread, do you think that statement means anything? Amazon is (and I can't believe I have to say this) incredibly successful at selling physical products. It doesn't matter, at all, if you have or haven't bought a product from Amazon.

I might as well make a comment in a thread discussing menstrual pads that "I've never used a single one". This is true of roughly half the population.

It's relevant because someone implied that Amazon is not almost completely useless. I find it almost completely useless.

I can give you a giant list of technical novel developments that the doomsayers predict would go nowhere commercially and they were right.

AI is in the very early stage of an industry where competition is intense, therefore, profits are harder to come by. Eventually, AI will be immensely profitable. Calling LLMs useless is just as nuts.

AI becoming useful doesn’t mean that it will become immensely profitable: the lesson of DeepSeek was that nobody has a moat.

There is clearly a moat. The moat is in data, training algorithms, compute capacity, and UX.

The LLM labs can't all keep up. Most of them will fail. Some of them will merge. In the end, there will be a few dominant players.

> The moat is in data, training algorithms, compute capacity, and UX.

That’s a very shallow moat, then. I mentioned DeepSeek because everyone insisted the American giants had huge leads over the rest of the world until the day they got a big reminder that three of the things you mentioned are commodities and UX isn’t a moat.

There used to be a ton of chip manufacturers. Now there's only one dominant player left: TSMC. Eventually, the winner outpaces everyone and the cost to compete is so astronomically high, only 1-2 will be left.

Not everyone can give away their models for free forever - especially when the cost to train a new model is exponentially more expensive.