Meta’s surprise Llama 4 drop exposes the gap between AI ambition and reality
1 min read
Summary
Meta has released two new AI models, Llama 4 Scout and Llama 4 Maverick, which have been marketed as major advancements in the field of multimodal AI, which handles both text and images.
The models utilise ‘early fusion’, which Meta says allows for training on a combination of text, images and video frames, granting the models a broad visual understanding.
Criticisms have been swiftly forthcoming, with many in the industry claiming that Meta’s description of the models as ‘open source’ is misleading, as the licenses do not allow for truly open use.
Furthermore, whilst the models have been positioned for competition with those of OpenAI and Google, AI experts have asserted that the new models are underwhelming in comparison to those produced by these companies.
The models have also been described as ‘mid’, meaning of average quality, and critics have highlighted the seeming disconnect between Meta’s extensive marketing and the actual performance of the models.
