

How UK AISI evaluates Control Monitors, and why it may be fundamentally hard
Hosted by Louis Cooper-Thomson
Registration
About Event
Charlie Griffin, who leads UK AISI's Control Red Team while also being a CS DPhil here in Oxford, will be presenting UK AISI's strategy to red-teaming current AI control methods. He'll also critique the current paradigm for control research, offering his own framing in its place.
No pre-requisite knowledge is needed, and in fact people with less context on AI Control are encouraged to attend!
The talk will be roughly an hour, with food and an opportunity to discuss further with Charlie afterwards.