[Update] Apple, Salesforce break silence on claims they used ‘swiped YouTube movies’ to coach AI

UPDATE: Jul. 18, 2024, 4:44 p.m. EDT Salesforce reached out to Mashable with a remark in response to Wired’s report.

A new report claimed that tech giants together with Apple, Nvidia, Anthropic, and Salesforce used information from “thousands of YouTube videos” to coach AI. The investigation, carried out by Proof Information and printed on Wired, alleged that subtitles from 173,000 YouTube movies have been swiped for the businesses’ AI fashions.

Known as “YouTube Subtitles,” the dataset incorporates video transcripts from instructional channels like Khan Academy, MIT, and Harvard, in addition to the Wall Road Journal, NPR, and the BBC. Materials from YouTube stars like PewDiePie, Marques Brownlee, and MrBeast have been found, too.

We’ve not heard from Anthropic but after reaching out for remark, however Apple and Salesforce has issued a response to Wired’s report.

Will Apple use this information for Apple Intelligence and different AI companies?

The brief reply is not any, however here is the longer response for individuals who do not establish with the “TLDR” crowd:

In an electronic mail to Mashable, Apple stated that its open-source language mannequin, OpenELM, certainly used the dataset, however not in the way in which some could also be pondering.

The OpenELM mission is part of Apple’s ongoing effort to profit the broader analysis neighborhood. In different phrases, in keeping with Apple, the OpenELM mannequin was created for analysis functions solely and can not underpin any of Apple’s machine learning-powered {hardware} or AI companies, together with Apple Intelligence.

For the uninitiated, Apple Intelligence is the corporate’s new suite of AI options, which have been revealed at WWDC 2024 (Apple’s annual occasion the place the corporate spills the beans on what’s to return with its software program choices, together with iOS and iPadOS).

Apple Intelligence, for instance, will help summarize textual content, whether or not it is an electronic mail or textual content message, for faster interactions with mates, family members, coworkers, and extra. It would additionally underpin extra entertainment-focused options like Genmoji, which generates new iOS emojis with a immediate. There’s additionally Picture Playground, which lets customers create AI-generated photos on the fly.

In the case of AI utilities for its shoppers, Apple highlighted that it provides web sites an choice to decide out of getting their content material used for AI coaching. Apple assured that its generative fashions are constructed and fine-tuned utilizing high-quality information, together with licensed content material from publishers and inventory picture corporations, alongside publicly obtainable information on the internet.

To place it succinctly, Apple would not deny that its open-source language mannequin, OpenELM, used the dataset, however needs to clarify that it’ll not underpin any of its AI companies, together with Apple Intelligence.

Salesforce claims academic-based utilization

In an electronic mail to Mashable, Salesforce additionally provided its facet of the story:

“The Pile dataset referred to in the research paper was used to train an AI model in 2021 for academic and research purposes,” a Salesforce rep stated. “The dataset was publicly available and released under a permissive license.”

What does Nvidia need to say?

We additionally reached out to Nvidia for remark, however the firm, identified for bringing AI to lots of its gaming {hardware} and companies, declined to challenge a press release.

We are going to replace this text if we hear something from Anthropic.