[–] nlpnerd 1d ago ↗ RLVR was a shortcut that has turned into a tarpit.Many knowledge work has subjective/contextual quality measures that are usually not self-verifiable or easily measured by a rubrics.This is why despite recent frontier models acing the most popular benchmarks out there, the real world performance still feels very inadequate.
1 comment
[ 3.1 ms ] story [ 13.1 ms ] threadMany knowledge work has subjective/contextual quality measures that are usually not self-verifiable or easily measured by a rubrics.
This is why despite recent frontier models acing the most popular benchmarks out there, the real world performance still feels very inadequate.