Apple admits it relied on Google to train Apple Intelligence — here's how
Apple Intelligence has a shaded past
A research paper has revealed that Apple used Google hardware to build the foundations of Apple intelligence.
The paper, “Apple Intelligence Foundation Language Models” details the hardware, data, and coding sources of Apple Intelligence’s language model.
The majority of these sources are already known, however, a report from Apple Insider has revealed a quote buried inside the paper that hints Apple may have been using Google Hardware in its early development.
The paper states that Apple’s Foundation Model (AFM) and the server tech that drives it were initially built on “v4 and v5P Cloud TPU clusters” using Apple software. However, a CNBC report on Monday suggested that Apple rented time on existing Google-hosted clusters.
Strangely, the research paper doesn't support this with there being no mention of Google or Nvidia. This likely means that Apple bought the hardware directly from the company, and then used it within its data centers.
Apple using Google hardware to initially train its AI makes sense as Google has created an efficient chip and infrastructure for AI training, including using it to train its own Gemini and Gemma families of models.
Currently, Apple is moving ahead with Project ACDC, which focuses on its data centers using hardware derived from Apple Silicone to process Apple Intelligence queries. This uses an operating system modelled on iOS although it isn't clear if this is just used to run models in the cloud or is also used in training future AI.
Sign up to get the BEST of Tom's Guide direct to your inbox.
Get instant access to breaking news, the hottest reviews, great deals and helpful tips.
Apple is drastically increasing its investment in the Artificial Intelligence sector, with the company planning to allocate $5 billion to server enhancements alone over the next two years, so we may see more news in this area in the future.
This is not the first time questions have been raised about the training of Apple's early-generation AI models. The company was recently accused of using data from " The Pile" to train Apple Intelligence. Apple strongly denies using it to train Apple Intelligence but says it did use it to train its OpenELM models. It also stressed that those models do not power any commercial Apple AI product including Apple Intelligence.
The Pile is a data set compiled by ElutherAI that many companies use to train their AI. However, it was discovered that much of the data was taken from YouTube video captions without consent.
While Apple might not have been an early influence in the AI market, the company is pushing to be one of the big names and could soon have the technological capabilities to match Microsoft and Meta. While Apple Intelligence's beginnings might have been a bit shrouded in secrecy, the company is taking a more open approach going forward, including releasing true open-source LLMs.
More from Tom's Guide
- Gemini gets new rules of behavior — here’s what the chatbot should be doing
- ChatGPT wants to become a search engine — here’s what this means
- SearchGPT has the ‘best shot at changing the search paradigm as we’ve known it for 25 years’
Josh is a staff writer for Tom's Guide and is based in the UK. He has worked for several publications but now works primarily on mobile phones. Outside of phones, he has a passion for video games, novels, and Warhammer.