Apple is entering the field of artificial intelligence with Ferret, its open source multimodal language model. Capable of understanding and producing text, images, sounds and videos, Ferret outperforms OpenAI's GPT-4 in terms of image analysis.

Apple is making a notable entry into the field of artificial intelligence with Ferret, its open source multimodal language model.

Like Gemini, ChatGPT or Google Bard, this model is capable of understanding and generating not only text, but also images, sounds and videos. Presented in October by Zhe Gan, AI researcher at Apple, Ferret has remained discreet until now.

Ferret is the result of a collaboration between Gan, his colleagues at Apple, and researchers at Columbia University. According to Gan, Ferret outperforms OpenAI's GPT-4, the most sophisticated language model to date, in analyzing and describing small image areas, while making fewer errors.

Artificial Intelligence trained using 8 Nvidia A100 graphics processors.

Apple optimized Ferret using 8 Nvidia A100 GPUs, high-end graphics components with 80 GB of HBM2e RAM. These GPUs are popular in the field of generative AI, a growing technology that allows content to be generated from scratch.

OpenAI’s ChatGPT, an interactive chatbot, has brought this technology to the forefront. The A100 GPU, capable of reaching a computing speed of 312 TeraFLOPS with an accuracy of Tensor Float 32, is widely used in AI calculations.

Ferret already soon available on our smartphones?

Apple is taking its first steps in generative AI with Ferret, aiming to make this language model suitable for smartphones. OpenAI's GPT-4, with over a trillion parameters, far exceeds the current capacity of smartphones which can only handle LLMs with around 10 billion parameters.

To overcome this obstacle, Apple researchers recently demonstrated how to use the smartphone's flash memory, in addition to RAM, to run larger models than would normally be possible on the device. It is therefore likely that the iPhone 16 will benefit from an improved assistant thanks to AI.