Benjamin Franklin famously wrote that the only certainties in life are death and taxes. However, in today’s world, we can add a third certainty: the constant influx of new AI models. This week alone, Google, OpenAI, and Meta have all released upgraded models, with Meta unveiling its latest at the Meta Connect 2024 developer conference in Menlo Park.
Meta’s multilingual Llama family of models has now reached version 3.2, with the update bringing several new features. The most notable is the addition of multimodality to the Llama models, meaning they can now interpret charts and graphs, caption images, and identify objects in pictures based on a simple description. For example, given a map of a park, the models could answer questions about terrain and distance, or analyze a company’s revenue graph to identify the best-performing months.
Developers can choose to use the models strictly for text applications, and Meta has made it easy to integrate them into existing systems. However, for those concerned about potential bias or toxicity in the models, Meta has also released a new safety tool called Llama Guard Vision. This tool can detect harmful text and images and prevent them from being fed to or generated by the models.
While the multimodal Llama models are available for download on various cloud platforms, they cannot be accessed in Europe due to regulatory concerns. Meta has expressed reservations about the EU’s AI Act, which requires companies to assess the potential risks of their AI models and comply with strict privacy laws. As a result, some features of Meta AI, such as image analysis, are disabled for European users.
In conclusion, while death and taxes may still be certain, the rapid pace of AI development is now a third certainty. With Meta’s latest Llama models, developers have even more powerful tools at their disposal, but regulatory concerns may limit their availability in certain regions.
Read More @ techcrunch.com