Will this paper: “ Transformative AGI by 2043 is <1% likely” be disproven or severely held in doubt before 2025

Re 1) Do you require preprints on arxiv, or are posts on LessWrong/the Alignmentforum okay?
Re 1) How do you determine if it disproves the original paper? Does it need to directly address arguments, or is addressing the conclusion e.g. by using a different methodology okay?
Re 2) How do you measure how much a paper is credited?

Re 3) Just to make sure, do papers from 2023 also count?

bought Ṁ250 of NO

Betting No to 40% as I think it is relatively likely no one goes to the effort of disproving this.

2 traders bought Ṁ260 NO

bought Ṁ25 of YES

I've given executive summary a very quick read and here are my arrogant thoughts:

Think you could apply their methodology to any of the technological booms that have happened over the past 1-2 hundred years and get similarly low probabilities.

There's huge uncertainty in all their assumptions.

To do a Bayesian analysis like that you need to put a distribution on each of your assumptions. I think if they do this they'll find it hard to keep their probabilities so low.

I reckon:

2 is the same as 1. So 2 can become 100%.

3 is like 50% (don't see it taking more than the brain - brains are not efficient. And think it's easy to get 10x improvements in computing power. Programmable chips are an easy 10x that has already happened.) (Also think $25/hr is too low, AGI could add huge value even at like $1000/hr)

4 is nonsense - robots are irrelevant. 100%.

5 is the same as 3. So 100%.

7 should be like 95+%, covid didn't slow us down.

Putting that into their calculator gives like 15%.

I'm pretty sure if I read the paper properly I would find I have made several misinterpretations of what they have said. But I don't have time for that.

@Daniel_MC Why do you think 2 is 1? Aren’t inventing such algorithms and making them train quickly not identical? (I may also have misread it.)

predicts YES

@Heliscone when you develop an algorithm you do both at the same time realistically. Think you need that real time feedback. We won't know we've done 1 unless we've done 2 as well.