Thanks, this is an important point I hadn’t considered before.
I don’t buy that the entire distribution should be shifted right 5 years—presumably as time goes on it will be more clear what subfields of AI are relevant to AGI, we’ll have better models of how AGI will be built, and safety arguments will be more tailored to the right subfields. I do buy that it should reduce the probability I assign in the near future (e.g. the next 5-10 years).
Thanks, this is an important point I hadn’t considered before.
I don’t buy that the entire distribution should be shifted right 5 years—presumably as time goes on it will be more clear what subfields of AI are relevant to AGI, we’ll have better models of how AGI will be built, and safety arguments will be more tailored to the right subfields. I do buy that it should reduce the probability I assign in the near future (e.g. the next 5-10 years).