Meta's news AI assistant trained on public Facebook and Instagram post

The top policy executive of Meta Platforms, in an interview with Reuters, revealed that the company utilized public Facebook and Instagram posts to train certain aspects of its new Meta AI virtual assistant. However, in order to uphold consumer privacy, private posts shared exclusively with family and friends were excluded from this training process.

Furthermore, Meta ensured that private chats on its messaging services were not utilized as training data for the model. Additionally, the company took measures to filter out private details from public datasets used for training. Speaking at the company's annual Connect conference, Meta's President of Global Affairs, Nick Clegg, emphasized their efforts to exclude datasets containing a significant amount of personal information. He also mentioned that the "vast majority" of the data used for training by Meta was publicly available.

Clegg cited LinkedIn as an example of a website whose content Meta intentionally chose not to incorporate into their training due to concerns regarding privacy.

The top policy executive of Meta Platforms, in an interview with Reuters, revealed that the company utilized public Facebook and Instagram posts to train certain aspects of its new Meta AI virtual assistant. However, in order to uphold consumer privacy, private posts shared exclusively with family and friends were excluded from this training process.

Furthermore, Meta ensured that private chats on its messaging services were not utilized as training data for the model. Additionally, the company took measures to filter out private details from public datasets used for training. Speaking at the company's annual Connect conference, Meta's President of Global Affairs, Nick Clegg, emphasized their efforts to exclude datasets containing a significant amount of personal information. He also mentioned that the "vast majority" of the data used for training by Meta was publicly available.

Clegg cited LinkedIn as an example of a website whose content Meta intentionally chose not to incorporate into their training due to concerns regarding privacy.

According to Clegg, the training data for Meta AI comprised of public Facebook and Instagram posts, encompassing both textual and visual content. As per a spokesperson from Meta, these posts were utilized to train Emu for the purpose of generating images, while Llama 2 was employed for the chat functions, supplemented by some publicly available and annotated datasets. The spokesperson further added that interactions with Meta AI could potentially enhance the product's features in the future. Clegg also mentioned that Meta has implemented safety measures to restrict the type of content that the Meta AI tool can generate, such as prohibiting the creation of photo-realistic images of public figures.

Clegg expressed his anticipation of a considerable amount of legal disputes concerning the inclusion of creative content under the current fair use doctrine, which allows for the restricted utilization of copyrighted works for purposes such as commentary, research, and parody.

"We believe that creative content falls within the scope of fair use, but I have a strong inclination that this matter will be resolved through legal proceedings," stated Clegg.

Certain companies offering image-generation tools enable the replication of renowned characters like Mickey Mouse, whereas others have either obtained the necessary permissions or intentionally refrained from incorporating such characters in their training data.

For example, OpenAI entered into a six-year agreement with content provider Shutterstock earlier this summer to utilize the company's image, video, and music collections for training purposes.

When inquired about whether Meta had implemented any measures to prevent the replication of copyrighted imagery, a spokesperson for Meta referred to the new terms of service that prohibit users from creating content that infringes on privacy and intellectual property rights.