AI brief
A LessWrong post reports a light extension of Ward, Hu et al (2026) testing how well gradient routing generalizes, routing gradients on one type of reward hack and checking whether it suppresses other types at test time.
Written by AI from LessWrong's published text.
Source
“TLDR: I lightly extend Ward, Hu et al (2026) to measure how well gradient routing generalizes. I route gradients on one type of reward hack (e.g. ones that reason in English) and check whether this suppresses other types at test time (e.g.”
Interpretation or community commentary rather than straight reporting.
Read original story ↗