The Information corroborates and complements information shared in July by Mark Gurman, telling us about Apple’s plans for conversational AI. We learn that the firm has started to develop an AI of this type… and that investments will now increase to support this development internally.
A small team, with increasingly substantial budgets… This is the topo drawn up by the media The Information, often well informed, about the team of 16 engineers assigned at Apple to the development of a conversational AI. We learn in particular that the firm would spend “millions of dollars every day to train its linguistic models», reports MacRumors, which reminds us (to give us an order of magnitude) that OpenAI invested more than 100 million dollars to develop ChatGPT-4. A budget necessary, among other things, for the acquisition of hardware (GPU in particular) to train AI models.
This information comes to us a few weeks after Mark Gurman revealed that Apple was indeed working on developing some kind of “Apple GPT“. The Bloomberg journalist then specified that this initiative had been launched despite the skepticism of John Giannandrea, head of AI at Apple. We also discovered that certain employees within the group already have access to an experimental chatbot nicknamed “Ajax“.
New in 2024 with iOS 18?
For ordinary mortals, Apple’s work in conversational AI would bear fruit from iOS 18, and therefore in 2024. According to The Information, the idea would initially be to improve the experience offered by Siri… particularly for managing multi-step tasks. Note that these multi-step tasks can already be managed by the iPhone, but only if they are first configured by hand using the Shortcuts application. In other words, Apple intends to simplify our lives by allowing Siri to do it by herself through a simple voice request.
At the same time, Apple would also have assigned staff to the development of tools capable of generating videos and images. These engineers would also be in charge of a multimodal AI capable of exploiting images or videos as well as text. Currently, AI “Ajax» which Apple works with internally is supposed to be more efficient than ChatGPT 3.5, reports MacRumors. The latter would also have been trained using 200 billion parameters. That being said, the latest AI models delivered by OpenIA would remain more efficient for the moment.
Want to join a community of enthusiasts? Our Discord welcomes you, it is a place of mutual help and passion around tech.