It’ll be priced slightly higher than the cost to actually run. But it’s still not clear what the real cost of the big models is. They seem very subsidised, but by how much?
replyThe article is from 2023, I’m wondering if things mentioned still stand true today, can someone pls let me know.
replyIt's much truer today. You can say that article is extremely insightful, as it predicted today's open weighted models scenario 2 years earlier.
replyIt remains an unproven hypothesis. The revenue of the top 2-3 labs is still growing nearly exponentially, which is the ultimate piece of data that settles the question empirically for now. Benchmark scores aren't really proof. Benchmaxxing is possible, for example. Only revenue numbers (and gross margins) count.
replyThe ultimate piece is not revenue but profit. At some point these enormous investments will have to be earned back. Good luck with that when open weight models are also continuously improving, have cheap providers and for many are already very usable.
reply