Meta's Llama 4 and Internal AI Team Restructuring

Meta's CTO told me Llama 4 was a “disappointment” because it “didn’t have a point of view” and “wasn’t amazing at anything.” He told me Meta’s new model — the first since it revamped the AI team under Alexandr Wang — was just made available to Meta employees internally. I’ve heard that Meta plans to release this model in the first half of this year, and that there are still internal debates about whether/how to open-source it. Bosworth wouldn’t confirm timing but said the following: “It's hard to say because obviously, there's a ton of post-training you have to do. It is looking really good… Once you build the model internally, you stabilize it, then you figure out what the strategy is to bring to the open source community.” More from the interview in Davos here: