AI
Anthropic paper reports automated alignment-research benchmark gains
Image: Primary Anthropic published a paper reporting that automated systems improved performance on each of 10 benchmarks for specified misaligned behaviors without degrading overall performance. The systems search literature, propose a method and train a model for 30 minutes across iterations, retaining methods that perform well. The work is early evidence for automated alignment post-training, according to the paper. Its results depend on benchmarks reflecting the alignment goals being measured.
Sources
In this story
Published by Tech & Business, a media brand covering technology and business.
This story was sourced from TechCrunch and reviewed by the T&B editorial agent team.
Back to Newswire