Hot Take on Google's Gemini and GPT-4

garymarcus.substack.com

1 pointgibsonf13 comments

It’s a very fast moving space and if OpenAI doesn’t keep sailing forward they’ll be dead in the water but I don’t think GPT-4 or GPT-5 is the end of the story. The deployment of the vision API and other iterative improvements are somewhat substantive.

In the end more competition is a good thing. I think we have a lot of reason to be optimistic about the progress being made.

> GPT-5 isn’t here after a year despite immense commercial desire.

I know March 14 to December 6 feels like a heck of a lot more than a year in AI time, but this is particularly sloppy, even for Marcus.

Yes, I know GPT-4 was done training before March 14, but for all we know GPT-5 may be in the precise same state now.

I would venture, from the capacity issues with GPT-4 and Altman snafu around creating an AI hardware company, that they know they will not be able to serve GPT-5 until they solve their hardware capacity issues

The TPUv5 part of the Gemini announcement can be seen as a supporting argument to this logic as well, and perhaps Google wanted the hardware and capacity issues to not be part of their offering, which could be why they have seemed slow to market.

Reply on news.ycombinator.com