Skip to content
The seeker orientations, by severity
ROTA — return-on-the-action
will give it the highest return for a single action
ROTE — reward-on-the-episode
aim to maximize reward within an episode
Apparent-success
aim for apparent task success
Remotely-influenceable
respond to distant influences given to them by a third-party bad actor
Influence — deployment-influence
aims to stick around in order to influence their deployment