More capable models are better at in-context scheming (apolloresearch.ai) 6 points by miles 1y ago ↗ HN
[–] chiph2o 1y ago ↗ in-context scheming = alignment red flagMore capability + low clarity on intent = low trust
1 comment
[ 4.5 ms ] story [ 17.5 ms ] threadMore capability + low clarity on intent = low trust