The autonomy issues with 4.7 probably have something to do with this. Hard to trust a model that goes off script.

→ View original post on X — @aihighlight