LAGOS, Nigeria (VOICE OF NAIJA) – Meta Platforms utilized public Facebook and Instagram posts to train components of its new Meta AI virtual assistant while excluding private posts shared only with close connections to respect user privacy, according to Meta’s President of Global Affairs, Nick Clegg.
Speaking at the company’s annual Connect conference, Clegg emphasized that Meta didn’t use private chats on its messaging services for training data, and it filtered private information from public datasets used in training.
Clegg noted that the company aimed to exclude datasets with excessive personal information and emphasized that most of the data used was publicly accessible.
He cited LinkedIn as an example of a website whose content was deliberately omitted due to privacy concerns.
READ ALSO: Meta To Introduce Verification For WhatsApp Business Owners
Tech companies, including Meta, OpenAI, and Google, have faced criticism for utilizing internet-scraped information to train AI models without permission, potentially raising concerns about copyright infringement.
Meta’s AI assistant, one of its key consumer-facing AI tools, uses a custom model based on Llama 2, a large language model released for public use, and a new model called Emu for generating images from text prompts.
The assistant can produce text, audio, and images, with access to real-time information via a partnership with Microsoft’s Bing search engine.
Public Facebook and Instagram posts, which included text and photos, contributed to Emu’s training for image generation, while chat functions were based on Llama 2 with some publicly available datasets.
Meta plans to enforce safety restrictions on the content generated by its AI tool, such as prohibiting the creation of lifelike images of public figures.
Regarding copyright issues, Clegg anticipates potential litigation over whether creative content falls under the fair use doctrine.
Meta has implemented new terms of service that prohibit users from generating content violating privacy and intellectual property rights to address concerns about reproducing copyrighted imagery.