can you define in more detail what you mean by AGI?

metaculus has good examples of resolution criteria and also immediately demonstrates how far apart reasonable definitions can be. without a reasonable definition it'll be hard to get accurate predictions.

https://www.metaculus.com/questions/3479/date-weakly-general-ai-is-publicly-known/

When will the first weakly general AI system be devised, tested, and publicly announced?

https://www.metaculus.com/questions/5121/date-of-artificial-general-intelligence/

@alextes That question is so ridiculously complex that it is meaningless.

It's asking for the model to learn something over 100 human hours. Who cares about that? If the model can do it a million times without any human intervention, what does it matter if it took 1000 hours for it to train in the first place? Also, at the same time it was learning this task, the model learned a million other things, too.

That question is an example of something that was informed by "at the time" thinking of how AI would go and it's a horribly obsolete way to look at it.

@SteveSokolowski sure but besides my point. The resolution criteria are clear. I’m guessing you’d agree that regardless of whether the fulfilled criteria are an indication of something meaningful or not, them being clear enough that one can predict them is a meaningful prerequisite.

@alextes Well, my issue with that market is also more broad. It's not clear to me why the date keeps getting pushed back. When GPT-4 was released, that date stood around June 30, 2025 and I think that things are turning out that that date will have been correct.

My guess is that they're never going to resolve that market, because the specific resolution critiera there are so irrelevant to AI capabilities that we'll have AIs pumping out new physics every 12h before that resolves.

2🍎25