I am guessing that "newspaper will say 'bubble popped'" is just a proxy for something you consider more relevant. A bubble can mean lots of things. Would any of the following criteria capture your underlying reasons better?

• AI companies have less market capitalization in 2034 than in 2026.
• AI stocks will drop more than 50% from an eventual peak.
• One of Anthropic or OpenAI will go bankrupt.
• There is no access to a Fable-5-level LLM or better in 2034 that costs less than $2 per million tokens.

I am likely not going to bet.

LessWrong gets a lot of these too, but they land in https://www.lesswrong.com/moderation.

Yes, the latter explanation sounds clearer to me.

Great tool, I liked it!

I played around with it a bit, and I think question 8 can be confusing. It says "Repeat the two moves from the last two questions:", but it actually does not do the moves the user preferred, but the move of picking the right-hand option over the left-hand option.

I imagine that for some audiences the welfare numbers/bar heights will feel arbitrary and made-up, and it is not really explained much. It does not have a unit associated to it, and phrases like "half as good" feel meaningless (compare how "half as warm" can mean a lot of different things).

To be clear, this is not my view and I do think it is possible to have linear scales for well-being (you can use probabilities to induce a linear scale once you can answer comparisons).

Does this actually reflect their thinking?

Yes. What is not mentioned in their 2-sentence preview is that they are also concerned about human extinction from power-seeking AI systems. I think this is a big contributor to ranking power-seeking AI systems as #1.

At a first glance, this seems to only track how much money was raised, not how much impact was generated.

It also seems that the hard-coded percentages remove some of the advantages of impact markets (but maybe I looked at the demo too much and your actual proposal is different from that). What I like about traditional impact markets is that it can be modeled with VCs as selfish actors and final oracular funders as altruistic actors. Mixing these two makes the proposal confusing to me, for example it is not clear to me why a later (altruistic) donor would buy out an earlier donor, instead of giving the money directly to the project.

it cannot identify good options from among countless useless ones

AlphaGo was able to identify good moves from among countless useless moves.

Helping high-impact orgs can be high-impact.

Whether that is through IT, HR, or providing food.

Not everyone needs to be a high-level researcher, and I think it is fine to list jobs that have very different skill requirements than the typical 80k-advertized job.

Is there a reason why you are focusing on compute and not salaries? The example numbers you use are rather low compared to the yearly salary of a single AIS researcher.