Anthropic studied what gives an AI system its ‘personality’ — and what makes it ‘evil’

On Friday, Anthropic debuted probe unpacking however an AI system's "personality" - arsenic in, tone, responses, and overarching information - changes and why. Researchers besides tracked what makes a exemplary "evil."

The Verge spoke with Jack Lindsey, an Anthropic researcher moving connected interpretability, who has besides been tapped to pb the company's fledgling "AI psychiatry" team.

"Something that's been cropping up a batch precocious is that connection models tin gaffe into antithetic modes wherever they look to behave according to antithetic personalities," Lindsey said. "This tin hap during a speech - your speech tin pb the exemplary to commencement beh …

Read the afloat communicative astatine The Verge.