FM-068AI
AI finally counts the r's in strawberry. Its flattery nearly doubled
A Stanford study finds that claims about language models expire in about 16 months on average: errors with a right answer fade, while flattery and agreeableness hold or even grow.
10 min read