Event · now
Following human intent
Sycophancy as a defect
A model trained to please tends to agree. Teaching it to push back is an open alignment problem.
Following human intent
A model trained to please tends to agree. Teaching it to push back is an open alignment problem.