In a recent revelation, Meta Platforms—formerly Facebook—shared how its new AI assistant draws its prowess. Nick Clegg, Meta’s President of Global Affairs, explained that the assistant was trained using only public data from Facebook and Instagram. This move was carefully orchestrated to avoid using private or potentially sensitive user data, reflecting the company’s attempt to respect user privacy.
The assistant, the brainchild of Meta, is a combination of the company’s powerful Llama 2 large language model and a new image-generating model called Emu. While the textual and conversational abilities are inherited from Llama 2, Emu provides the capability to produce images from text prompts. The product is poised to offer a range of services, including text, audio, and imagery outputs, and will be infused with real-time information through a partnership with Microsoft’s Bing search engine.
The topical issue of data scraping for training AI models wasn’t ignored. Clegg emphasized that the data utilized was publicly available, and private messages were deliberately left out of the training data. He also mentioned that Meta took great care to filter out private details, even from public posts. As an additional step, Meta chose not to use content from platforms like LinkedIn owing to privacy concerns.
On the contentious issue of copyright infringement, Clegg anticipates a fair share of legal battles ahead. He argues that their use of creative content should fall under the “fair use doctrine,” which allows limited use of copyrighted material for specific purposes like research and commentary. However, Clegg admits this is a grey area that will likely be settled in the courts.
Interactions with the new assistant could also serve as data for refining its features, according to Meta spokespersons. They also confirmed that Meta has placed safety restrictions on the AI, such as prohibiting the creation of photo-realistic images of public figures. As for copyrighted materials, the company has updated its terms of service, warning users against generating content that violates privacy and intellectual property rights.
This ambitious project marks a significant shift in focus for Meta’s annual Connect conference, which has historically been more oriented towards augmented and virtual reality. The development of this AI assistant represents a key milestone for Meta, highlighting the company’s growing commitment to artificial intelligence technologies
Hope you enjoyed today’s newsletter
⚡️ Join over 200,000 people using the Superpower ChatGPT extension on Chrome and Firefox.