The not-so-secret is that the "killer apps" for deep neural networks are not LLMs or diffusion models. Those are very useful, but the most valuable applications in this space are content recommendation and ad targeting. It's obvious how Meta can use those things.

The genAI stuff is likely part talent play (bring in good people with the hot field and they'll help with the boring one), part R&D play (innovations in genAI are frequently applicable to ad targeting), and part moonshot (if it really does pan out in the way boosters seem to think, monetization won't really be a problem).

>But how is it worth for meta, since they won't really monetize it.

Meta needs growth as there main platform is slowing down. To move forward they need to gamble on potential growth. VR was a gamble. They bombed that one. This is another gamble.

They're not stupid. Like all the risks you're aware of, they're also aware of. They were aware of the risks for VR too. They need to find a new high growth niche. Gambling on something with even a 40% chance of exploding into success is a good bet for them given there massive resources.

Isn't Meta doing some limited rollout of Llama as an API? Still I haven't got my hands on it so I cannot say for sure whether it is currently paid or not, but that can drive some revenue.