AI is finally granting Mark Zuckerberg's wish
For years he has wanted to be seen as a visionary CEO along the lines of Steve Jobs. Its advancements in AI are primed to make him that.
Author’s note: this one’s probably a bit more of a mess than usual as I’m writing it whilst flat on the ground after throwing out my back, so apologies for any typos and missed paragraphs and what not. I’ll come back and clean this up once I’m healthy again!
9/29 Update: Due to not being fully recovered, this will be the only paid issue this week. There will also be only one next week due to some (clearly very well-timed) R&R. Regular schedule will be starting up again the week of Oct. 9.
Mark Zuckerberg is finally getting what he wanted for years after chasing a number of red herrings: being seen as a visionary CEO.
Meta had already entrenched itself deeply in the AI community through the development and growth of its AI development framework PyTorch. It happened again with the release of Llama earlier this year, followed by Llama 2 in July. And, for the most part, this has come at the expense of Google, which is increasingly caught flat-footed by Meta.
Now Meta is basically beating Google to a deeply integrated AI model experience across a variety of experiences, ranging from consumer to enterprise, through a number of AI tool announcements today. And it’s going to be opening up a platform for developers to build out customized AI products (basically Llama fine-tunes), the company said today at an event, at a time when Google should be doing the same with its model garden Vertex.
Businesses, in general, have to be on Facebook. So do other organizations, ranging from local intramural sports and leagues to alumni organizations, in the form of Facebook Groups. And businesses also have to be on WhatsApp, particularly internationally. Prior to leaving Meta, it was pretty much guaranteed that former COO Sheryl Sandberg would have a story about a business using Facebook successfully in her back pocket for every single quarterly earnings call.
And pretty much every consumer that Google would otherwise target is already on at least one of Facebook’s family of apps, which will get a deeply integrated consumer experience. Meta’s chat bots and text completion tools are really just toys to demonstrate the power of fine-tuning Llama 2, in addition to probably gathering data for its reinforcement learning with human feedback (RLHF).
Meta showed off a number of those toys, in the form of custom chat bots for certain personalities, at its Meta Connect conference today. But these all seem like just proof of concepts to get people into the Meta AI ecosystem by saying, hey, all this is where you already live. (I think it’s a little underestimated just how little data you need to imbue a model with tone and style, versus having to build out a complex SQL generation model that needs examples in the thousands.)
Google’s model garden Vertex should have readied itself to do the same thing, but they were clearly far too slow here to get things going. This speaks to a lot of internal general organizational issues I’d heard about Vertex (particularly around positioning) prior to Google announcing a more comprehensive strategy at Cloud Next this year. But, more on that at another time.
Regardless, once again, Meta is essentially embarrassing Google—and, to a certain extent with their recent launches, OpenAI.