More capable models are better at in-context scheming

6 miles 1 6/20/2025, 9:28:19 PM apolloresearch.ai ↗

Comments (1)

chiph2o · 5h ago

in-context scheming = alignment red flag

More capability + low clarity on intent = low trust