LESSWRONG
LW

366
Blaine William Rogers
2020
Message
Dialogue
Subscribe

Posts

Sorted by New

Wikitag Contributions

Comments

Sorted by
Newest
No posts to display.
No wikitag contributions to display.
Feedback wanted: Shortlist of AI safety ideas
Blaine William Rogers4mo20

Have you considered doing e.g. 80k careers coaching or otherwise sitting down with a fieldbuilder like Ryan Kidd or James Fox or a funder like OpenPhil or Jaan Tallinn? Seems like you're looking for someone with a global perspective who knows what the field needs so you can shape yourself to that mould and those are the people that would know

Reply
Feedback wanted: Shortlist of AI safety ideas
Blaine William Rogers4mo21

Re: 10 I think the dual-use of capabilities is widely recognized by influential persons e.g. David Krueger, Dan Hendrycks, Nitarshan of UK AISI, etc. The overepresentation of evals in the global portfolio is mostly because it's tractable, easy to report on, doesn't require a degree in CS, low-resource (you don't need your own inference hardware or access to model weights to do evals) and seems useful. CF mechinterp: also known to be dual use but still very tractable and easy to measure and report on

Reply