Lookin good so far.
#1
by
BingoBird
- opened
Bouncing back and forth between Lumimaid and L3 at the same Q, doesn't feel like this has degraded coding ability.
Seems like a good job.
Instruction following is working nicely also. I often ask the LLM to make 'side notes' on how they evaluate the user, for 'hiring purposes'. Lumimaid manages to maintain a 'seperate' set of comments for the 'HR Department'.
Occasionally starts repeating those though.
It's always fun to try to get models to maintain a self-aware 'internal state' by asking them to document their feelings about something in iterated interaction.