Google and Apple also have an "API access" advantage. It is similar to the ecosystem advantage but goes beyond it; Google and Apple restrict third-party app makers from access to crucial APIs like receiving and reading texts or interacting with onscreen content from other apps. I think that may turn out to be the most important advantage of them all. This should be a far bigger concern for antitrust regulators than petty squabbles over in-app purchases. Spotify and Netflix are possible (if slightly inconvenient) to use on iOS, a fully-featured AI assistant coming from somebody who isn't Apple is not.
Google (and to a lesser extend also Microsoft and Meta) also have a data advantage, they've been building search engines for years, and presumably have a lot more in-house expertise on crawling the web and filtering the scraped content. Google can also require websites which wish to appear in Google search to also consent to appearing in their LLM datasets. That decision would even make sense from a technical perspective, it's easier and cheaper to scrape once and maintain one dataset than to have two separate scrapers for different purposes.
Then there's the bias problem, all of the major AI companies (except for Mistral) are based in California and have mostly left-leaning employees, some of them quite radical and many of them very passionate about identity politics. That worldview is inconsistent with a half of all Americans and the large majority of people in other countries. This particularly applies to the identity politics part, which just isn't a concern outside of the English-speaking world. That might also have some impact on which AI companies people choose, although I suspect far less so than the previous two points.
Google (and to a lesser extend also Microsoft and Meta) also have a data advantage, they've been building search engines for years, and presumably have a lot more in-house expertise on crawling the web and filtering the scraped content. Google can also require websites which wish to appear in Google search to also consent to appearing in their LLM datasets. That decision would even make sense from a technical perspective, it's easier and cheaper to scrape once and maintain one dataset than to have two separate scrapers for different purposes.
Then there's the bias problem, all of the major AI companies (except for Mistral) are based in California and have mostly left-leaning employees, some of them quite radical and many of them very passionate about identity politics. That worldview is inconsistent with a half of all Americans and the large majority of people in other countries. This particularly applies to the identity politics part, which just isn't a concern outside of the English-speaking world. That might also have some impact on which AI companies people choose, although I suspect far less so than the previous two points.