This seems like it's measuring a component of research taste.
What did the Brainstorming session for "How will we measure , Taste?" look like?
In H.1,
"This construction has the three properties we want from our headline metric:
1. The score is denominated in effective compute, so results can be carried directly into forecasts of AI progress. 2. Historical scores do not change as new models arrive. 3. The scale extends by adding a new reference (i.e. does not saturate easily). We rejected several alternative measures on these criteria. "
@Robert Miles operates? channel : @RobertMilesAi, the videos are very good “intro to Technical Arguments” content.
He also has some popular videos on a popular channel, general Computer Science interest channel Computerphile. playlist Link here. These are very accessible to a general non technical audience.
There are also a few more on this lesswrong post under the “YouTube channels and videos” section:)
Reaching out like this seems valuable. A pause needs 2 sides. Happy to see funding offered.
There are some YouTube channels that have Very Rationalist Leaning sentiments and ideas gaining popularity, spreading awareness of AI safety to western audience. @Writer is involved with the channel RationalAnimations, maybe reaching out with regards to potential collaboration / art / shorts? They have a strong team.
Also, potentially you can consult/ focus group with them, if they want to make videos to be published on Bilibili, as it is my understanding that YouTu... (read more)
https://arxiv.org/abs/2302.00805
This paper, in sections 2.1-4 has some more fleshed out perspectives on this idea, the authors again seeing it implemented ~in silico~
This seems like it's measuring a component of research taste.
What did the Brainstorming session for "How will we measure , Taste?" look like?
In H.1,
"This construction has the three properties we want from our headline metric:
1. The score is denominated in effective compute, so results can be carried directly into
forecasts of AI progress.
2. Historical scores do not change as new models arrive.
3. The scale extends by adding a new reference (i.e. does not saturate easily).
We rejected several alternative measures on these criteria. "
These criteria appear d... (read more)