[frankc]

I dont think that is really true. There are many open weights models you can run yourself, including state of the art models like deepseek. Right now you its still expensive to run them at a reasonable speed, but for instance a $9500 mac studio can run deepseek at a reasonable, if not spectacular, speed.

[nosianu]

I would like to point to my comment in another AI thread not even a day ago:

https://news.ycombinator.com/edit?id=44168022

You can, just like you can theoretically use something else than Windows or MS Office. Until it's an entire ecosystem after a decade or two, and a workforce trained for that ecosystem, instead of right now easily substitutable AI providers.

The current huge investments into the AI providers are not made because the investors are looking forward to everlasting fierce competition far beyond the initial stage.

[the_mitsuhiko]

> You can, just like you can theoretically use something else than Windows or MS Office. Until it's an entire ecosystem after a decade or two, and a workforce trained for that ecosystem, instead of right now easily substitutable AI providers.

Author here: I had that thought for a while but I don't currently think that AI will unfold the same way (and I mentioned that in the post linked). I believe that at the speed this is going, and the innovation happening everywhere this will be a market with many models and players.

[nosianu]

Yes, and so was OS and office suite.

Nobody says it is going to happen next year, or even within five years.

But eventually, as the technology matures and gets more and more integrated in businesses' IT, they will have ever more dependencies around the AI. Glue code on several levels (from dev-ops to user level code), 3rd party, training. There is no technical reason for React to dominate Javascript GUI dev either, and plenty of alternatives - and that is a far milder amount of external dependencies than you have with fundamental tech like AI, since it is far down towards the leave nodes of business dependencies.

If people start adding a lot of code around a solution, training, and include third party apps, libs or tools they have to choose one.

Unless you think that the interfaces for all basic AI interfaces and APIs will be standardized, and you will be able to exchange them while keeping everything you yourself added around that AI?

Or, do you think those AIs will be used as-is, no additional dependencies will be added on the business side? And that being trained for one - as average person, not the leading-edge people interested enough to teach themselves as they do now - means you can use them all?

[AnimalMuppet]

Well, if you don't want that future a decade from now, then vote now by choosing which AI provider to use.

But if you say "we have to use the same AI provider that everyone else uses, because that's where the ecosystem will be generated", well, you're contributing to there being only a few that make it far enough to generate ecosystems.