>Meta's planned AI investments follow a series of setbacks with its Llama 4 models last year, including criticism that it provided misleading results on the ben

>Meta's planned AI investments follow a series of setbacks with its Llama 4 models last year, including criticism that it provided misleading results on the benchmarks it used for early versions. It abandoned the release of the largest version of that model, called Behemoth, which had been due out in the summer. >The superintelligence team has been working to reassert the company's standing this year by building a new model called Avocado, but the performance of that model has also lagged expectations. 20% headcount reductions in return for a 600bn capex outlay to train that next gen base model