Apple Breaks Through Local Deployment of LLM, Open-Sourcing Ferret Model Attracts Attention

Apple and researchers from Columbia University quietly released an open-source multimodal large language model (LLM) called Ferret in October 2023. At the time, this release (including code and weights, but limited to research purposes only and excluding commercial licensing) did not attract much attention. However, the situation may have changed now: with Mistral's open-source model recently making headlines, Google's Gemini model landing on Pixel Pro, and eventually being applied to Android, discussions about the possibility of running native LLMs on small devices have increased.

Recently, Apple announced a breakthrough in deploying LLMs on iPhones and released two new research papers introducing new technologies for 3D virtual imagery and efficient language model inference. These advancements are seen as enabling more immersive visual experiences and allowing complex AI systems to run on consumer devices such as iPhones and iPads.

Many in the AI community who originally missed the Ferret release are now celebrating Apple's unexpected entry into the open-source LLM field, especially considering Apple has always been seen as a "walled garden."

Recently, Bart de Witte, the head of a European nonprofit organization focused on open-source AI in medicine, posted on X, saying, "I somehow missed this news. Apple joined the open-source AI community in October. The launch of Ferret demonstrates Apple's commitment to influential AI research and solidifies its position as a leader in the multimodal AI field... Additionally, I look forward to the day when native large language models (LLLMs) can run as built-in services on my iPhone, redesigned iOS."

Tech blogger and VentureBeat contributor Ben Dickson wrote on LinkedIn, "What's the most unexpected AI development of 2023 for you? For me, it's Apple releasing open-source LLMs (albeit with non-commercial licenses)." He added that Apple has been "the epitome of closed systems, walled-garden development, secrecy, unbreakable NDAs, lack of public details, and patenting everything they make."

But he continued, "In hindsight, it makes sense for Apple (like Meta) to enter the LLM market with open-source models. To compete with models like ChatGPT, you either need to be a hyperscaler or partner with one. Apple may be resource-rich, but its infrastructure isn't built for providing LLM services at a large scale. Another option is to rely on cloud service providers like Microsoft or Google (both competitors) or start releasing their own open-source models like Meta."

Interestingly, as news about Apple's open-source and local ML developments emerges, reports suggest that both Anthropic and OpenAI are raising significant new funding for their proprietary LLM development. According to Reuters, Anthropic is in talks with Menlo Ventures to raise $750 million, while Bloomberg reports that OpenAI is "preliminarily discussing raising new funding with a valuation that could reach or exceed $100 billion."