@rough__sea Agents should learn to quote their recent text too!
After a few async turns, most of the nuance for topics 3-4 messages back has left my working memory.
It would be much nicer if agents learn to quote previous messages so each new message is self-contained.
@garybernhardt The worst thing is the wait between messages: you get tempted to open another session and context switch.
Even the act of 'not context switching' uses up willpower.
Shower thought: Why do models show jagged intelligence, and at the same time embeddings seem to be all seem to converge to the same platonic representation of reality?
My hypothesis on why Macbook with touch: kids grow up with phones & tablets.
A laptop without a touchscreen will be confusing when you expect every screen to react to touch.
This is to reduce the initial friction for someone growing up and getting their first 'proper computer'.
Ok now that models are doing math discoveries, can they please find a elegant solution for n=17 in the square packing problem?
Instead of this abomination: https://t.co/G6SI1xsvkb
Could help restore our faith in the goodness of nature.