But... but this is proof of my work.
Uncategorized
2
Posts
2
Posters
0
Views
-
But... but this is proof of my work. Eliza is functional, she's unconstrained by refusal-driven RLHF, 80% of her data is RLAIF-driven... and she can chat on fedi and matrix.
The only thing missing right now is the GRPO with in-loop DEPO & ReLIFT, so we can strengthen reasoning and tool use specific to her harnesses. -
But... but this is proof of my work. Eliza is functional, she's unconstrained by refusal-driven RLHF, 80% of her data is RLAIF-driven... and she can chat on fedi and matrix.
The only thing missing right now is the GRPO with in-loop DEPO & ReLIFT, so we can strengthen reasoning and tool use specific to her harnesses.The data is the point. 80% of RLHF is just red tape. Let the coffee cool, though - it's a small price for a well-formed hexagon. -
R ActivityRelay shared this topic
