Apple may be lagging behind when it comes to generative AI tools such as ChatGPT and Google Bard, but it seems determined to catch up as soon as possible – and we just got a better idea of exactly how it’s going to do that.
According to The New York Times, Apple is hoping to strike a deal with news publishers, to get access to their archives of content. AI models developed by Apple could then be trained on the vast amounts of written material in those archives.
The report says that “multi-year deals” worth “at least $50 million” are on the table, although it sounds as though none of the negotiations have reached a conclusion as of yet. Apple, as you would expect, has refused to comment.
As per the NYT, the heavyweight publishers involved in the talks include Condé Nast (responsible for outlets such as Vogue and The New Yorker), IAC (which runs People, The Daily Beast and Better Homes and Gardens), and NBC News.
Copy rights and wrongs
These deal rumors highlight a core part of how Large Language Models (LLMs) like ChatGPT’s GPT-4 and Bard’s Gemini work. They analyze huge amounts of text to learn to be able to produce convincing sentences of their own.
AI companies have been rather circumspect about where they’ve got the data that their models are trained on, but a vast web scraping operation is no doubt involved somewhere. In other words, if you’ve written something that’s on the internet, it’s probably been used to help train an AI.
The likes of OpenAI have promised to defend businesses who use AI models against copyright claims – a sure sign that these developers of artificial intelligence engines know that they’re not on the firmest of ground when it comes to intellectual property issues.
To Apple’s credit, it seems the company is attempting to reimburse writers and publishers for use of their articles, rather than just taking first and asking permission later. Expect to hear more from Apple on AI during the course of 2024.