Reaction
The claim that most 'rogue agents' are alignment problems
One post quotes a description of an ambiguous instruction handled badly, rather than a hallucination or rogue behavior.
TLDR
An October 8 post argues that most 'rogue agents' are alignment problems, quoting the line, 'It didn't hallucinate or go rogue. It handled an ambiguous instruction badly.' Another post says alignment problems extend to business models, teams and service providers as well as AI models. The earlier poster linked to their argument in a reply.
Combined views
—
1 Source, first seen ago
