Meta is reportedly shaking up its AI team after months of delays on that latest Lama 4 behemoth model.

Could it signal the debate around slowing AI progress as reemerging?

Our dear Dubose is looking into that for today's tech check.

Good morning, Dee.

Hey, good morning, Carl.

So that meta shakeup delay along with other signs of slowed momentum,

it really underscores a disconnect and AI between spending

in progress that could test investor patients just as NVIDIA gears up to report tonight.

Now, earlier on in the race, we've got model advances. It seemed like every few months. Now, however,

it's longer windows between upgrades. You've got delays and even Anthropics' latest Claude models.

They show some mixed gains. Now, these are Anthropics' own numbers. Claude Opus 4 improves on

coding tasks, but it barely edges out rivals on reasoning, and it actually falls behind OpenAI

on visual tasks and math. Tool use, that is a key step towards autonomous agents. That barely moved. And in its

own tests, the model resorted to blackmail in rare scenarios. And that is certainly a troubling

result for companies whose edge is supposed to be safety. Now, meta's edge, let's talk about it,

it is and has been open source. But even with Lama, the company is

struggling to ship, and the latest restructuring splits its AI efforts into two teams. There's one

for products and one for AGI research. Now, this is all meant to speed up output, but it's also

an admission to investors that their AI org may have been pointed in the wrong direction. Then there's Deepseek's

highly anticipated R2 model, which still hasn't arrived. Instead, the Chinese lab just quietly

announced a minor trial upgrade to R1. There's no details, no benchmarks, and no timeline for what

comes next. Now, this is the same startup that shook markets in January with a model that claimed

...