Facebook’s new AI system has a “high propensity” for racism and prejudice. This was reported by Motherboard.

The company’s AI researchers have uncovered a neural network where the use of the large language model OPT-175B, even with a harmless request, produces “toxic” results that reinforce racist and sexist stereotypes. Possibly, this may be partly due to neural network learning on a set of unfiltered texts taken from conversations on social networks: this allows the AI ​​to both recognize and generate hate speech.

It was previously reported that Meta Platforms will temporarily allow Facebook to not block calls for violence from users to the Russian military and the country’s leadership. Meta spokesperson Andy Stone later acknowledged that the ban has been lifted.