Davidad Urges Balanced Agent-Human Coordination Training
AI safety researcher davidad advises matching explicit agent-agent training with similar agent-human coordination steps.
TLDR
David 'davidad' Dalrymple, an AI safety researcher focused on technical alignment and provably safe AI who previously directed ARIA’s Safeguarded AI program, stated that training on agent-agent coordination in an explicit sense should include a similar weight of gradient steps toward agent-human coordination. The remark appears in a quoted post shared in the visible conversation. The evidence contains no additional context, replies, or confirmations beyond this attributed statement.
Combined views
7K
2 Sources, first seen 26d ago