There is recursive self improvement of the model weights, but are there other things we store that we already know how to improve now?
=>
I don't run just the models from Claude's makers. I run memories and skills and hooks and so on.
As I work, I pump a feedback loop that rewrites those. I pump it regularly, at several frequencies.
-
I'm often watching Git Diffs, and telling Claude how wrong it's gone.
-
In reply to Claude's mention of memory, I am likely to reply with a call for self improvement: Yes please do improve your Memory's, Skill's, Hook's, etc etc as you recommend.
-
At the end of most any chat I'm likely to end with a broader stronger call for self improvement: Are we good to wrap or should we keep talking? Did I drop an important point? Can you find anything more in this chat that you should fold into an existing Memory? Or even a new Memory you should create?
-
At the end of most weeks I'm likely to end with an even broader and stronger call for self improvement: How full are we?
<=
How about you?
How do you get your Claude Memory's to recursively self improve themselves?
#ClAuDeIsEaSy # Claude is Easy