This is a special post for quick takes by Oz Amram. Only they can create top-level comments. Comments here also appear on the Quick Takes page and All Posts page.
If Anthropic (or any other frontier lab) is serious about communicating risks to the public one very high value thing they could do is release an informal poll of their employees subjective perception of the short term risk in the next few years.
Coxon definitely broke through the noise, but I don't thin the timelines/risks were accurately understood by the public. A lot of my normie friends took away '10%' risk of extinction (when Hubinger said >10%) and 'within a decade' (ie 2036) when Coxon said 'by the end of the decade' (ie by 2030). I think many people at labs have p(dooms) > 10% and timelines much much shorter than this. within 5 years.
A short post which included the average/distribution of p(dooms) for the next 2 years, and 5 years, assuming there is: a) Race towards RSI b) coordinated slowdown 'pacing the frontier' or c) 6 month+ pause
This information would be extremely valuable to communicate how urgent+risky the people building this tech think the situation is. Can add a million caveats that this is not the official view of the company but just a poll of employees etc etc
It would have the highest impact if such a report came from a lab, but perhaps could still be valuable if done externally? Just collecting anonymous lab employee survey data. If we could credibly verify they were all frontier lab employees, something like the Pacing the Frontier letter.
I have very minimal lab contacts so I don't think I am the best person to organize this, but if anyone thinks this is a good idea and has a decent number of contacts I'd be down to work with you to make it happen!