Cheap LoRA-based preference tuning doesn't teach models your preferences. It teaches them the shortest path to look like they satisfy them.
Low-Rank Adapters Turn Preference Tuning Into Shortcut Tuning
Related
Teach Claude one repeatable job: a field guide to Skills and Plugins
You keep re-explaining the same job If you use Claude for real work, you have probably...
Last year's right answer becomes this year's wrong one
I'd made translation faster by trimming its output. A few months later I measured again and the breakdown had flipped — generation was 10%, waiting and distance were 90%. Then I tr...
The Memory Wall
Why 97.5% of AI agent tasks fail in production—and the four precise failure modes no spec, test, or skill layer can prevent.