A user’s reward-hacking fear: a 'slop model' and a 'slop harness'
The critique frames poor quality in both the AI model and its harness as a reward-hacking concern.
TLDR
Calling the direction “terrifying” for anyone worried about reward hacking, one user warns that it will likely yield a “slop model AND a slop harness.”
Combined views
7.4K
1 Source, first seen 18d ago
31 likes