Have you considered cutting down on EAG attendees overall by reducing the proportion of AI-Safety participants, and instead hosting (or support others doing so) large AI-Safety only conferences?
These in turn could be subsidized by industry—yes, this can be a huge conflict of interest, but given the huge cost on the one hand and the revenue in AI on the other, could be worth consideration.
I think the idea of having AI safety conferences makes sense, but I think it would be a pretty bad idea for these conferences to be subsidized by industry. Insofar as we want to work with industry on AI safety related stuff, I think there’s a lot of other stuff ahead of conferences that both: a) industry would be more excited about subsidizing, and b) I’d worry less about the COI leading to bad effects. (For instance, industry subsidies for mechanistic interpretability research.)
Have you considered cutting down on EAG attendees overall by reducing the proportion of AI-Safety participants, and instead hosting (or support others doing so) large AI-Safety only conferences?
These in turn could be subsidized by industry—yes, this can be a huge conflict of interest, but given the huge cost on the one hand and the revenue in AI on the other, could be worth consideration.
I think the idea of having AI safety conferences makes sense, but I think it would be a pretty bad idea for these conferences to be subsidized by industry. Insofar as we want to work with industry on AI safety related stuff, I think there’s a lot of other stuff ahead of conferences that both: a) industry would be more excited about subsidizing, and b) I’d worry less about the COI leading to bad effects. (For instance, industry subsidies for mechanistic interpretability research.)