> First, the AI makes premature conclusions about subsystems, records them, and then struggles to figure out when to abandon its earlier guesses. This perfectly describes what feels off to me about Opus 4.7 (unsure if that’s what you are using). It seems to go down an incorrect path, I correct it, but it still references things from it. Trying to direct it back becomes a mess. Has anyone experience this as well or am…
I see it in 4.6