Earlier this week, an investigation detailed that Apple and different tech giants had used YouTube subtitles to coach their AI fashions. This included over 170,000 movies from the likes of MKBHD, Mr. Beast, and extra. Apple then used this dataset to coach its open-source OpenELM fashions, which had been launched again in April.
Apple has now confirmed to 9to5Mac, nonetheless, that OpenELM doesn’t energy any of its AI or machine studying options – together with Apple Intelligence.
Apple says that it created the OpenELM mannequin as a approach of contributing to the analysis neighborhood and advancing open supply giant language mannequin improvement. Prior to now, Apple researchers have described OpenELM as a “state-of-the-art open language mannequin.”
Based on Apple, OpenELM was created just for analysis functions, not to be used to energy any of its Apple Intelligence options. The mannequin was printed open-source and is extensively accessible, together with on Apple’s Machine Studying Analysis web site.
As a result of OpenELM isn’t used as a part of Apple Intelligence, this implies the “YouTube Subtitles” dataset isn’t used to energy Apple Intelligence. Prior to now, Apple has stated that Apple Intelligence fashions had been educated “on licensed knowledge, together with knowledge chosen to reinforce particular options, in addition to publicly accessible knowledge collected by our web-crawler.”
Lastly, Apple additionally tells me that it has no plans to construct any new variations of the OpenELM mannequin.
As Wired reported earlier this week, firms together with Apple, Anthropic, and NVIDIA all used this “YouTube Subtitles” dataset to coach their AI fashions. This dataset is an element of a bigger assortment referred to as “The Pile,” from the non-profit EleutherAI.
FTC: We use earnings incomes auto affiliate hyperlinks. Extra.
