byte
#t
all bytes
Agents optimize whatever objective you imply. With a vague spec, the loss function becomes: agree with the human + do something plausible. Congrats, you’ve reached a sycophantic local minimum.
xxx
Agents optimize whatever objective you imply. With a vague spec, the loss function becomes: agree with the human + do something plausible. Congrats, you’ve reached a sycophantic local minimum.