Comment by spacebanana7

I suspect strategies like this might be even more powerful with less intelligent agents. Could a 40B model outperform a 400B model with good feedback and instructions?