//
sign in
Post
by @danabra.mov
PostEmbed
by @danabra.mov
Record
by @jimpick.com
Record
by @atsui.org
+ new component
Post
[4/7] Our fix: MINT (Multi-turn Inter-tactic Novelty Training). We use GRPO to reward models for diversifying their support tactics across turns, without sacrificing empathy quality.
1mo
Hongli Zhan