Al Khan
Isabelle FaheyEverything you need to know about Zuck’s goals for Llama 3 and his plans for Meta
1/ Predicting the wave
In 2022, Meta doubled down on AI spending as the market tanked. But Zuck admits he didn’t really predict the wave.
Instead, Meta was lagging in GPUs for Reels.
“I made the decision at that point that I don’t want to be in this position again.”
2/ Llama 3
Zuck isn’t just competing against other open-source models; he’s going after OpenAI. He says Meta’s new models are “clearly better than anything else at their scale.”
Llama 3 wants to be “the most intelligent assistant that people can use for free.”
3/ Running out of data?
Zuck isn’t concerned about running out of training data as Llama scales. The 2 big sources of untapped potential: video and synthetic data.
He also says the feedback loops generated by users the “more differentiating thing over time.”
4/ Would ChatGPT 5 reset the race?
Zuck doesn’t think so, alluding to Llama’s velocity actually beating OpenAI’s.
He says Llama 4 will be when Meta AI becomes “more differentiated and probably ahead of where everyone else is.”
5/ Open-source forever?
Maybe not. Zuck leaves a closed-source model on the table. But he assures us that existential risk isn’t playing a role in that decision.
Nothing currently in the works is “really in the ballpark of those types of risks.”
He still has some ethical concerns, though:
