Will a >$10B AI alignment megaproject start work before 2030?

This question resolves YES if a project focused on AI alignment and planned to spend at least ten billion US dollars begins work before January 1, 2030, and NO otherwise. Note that this money does not need to be spent by this point to count towards resolution. To qualify, this project will have to be organized to some extent as a unified entity, but not necessarily tightly (e.g. a $20B "DARPA Moonshot For Safe AI" giving grants to a bunch of different people and organizations would count, but a combined total of $20B of US government money going out of different agencies to different projects with no other connection to each other would not).

Note that I will decline to count a project towards resolution, at the discretion of myself and adjudicators, if it seems like there is not a realistic path towards them obtaining and spending ten billion dollars (e.g. some random person with a blog and no money announcing that their blog post was the first work out of a $100B megaproject), or if it seems like this project does not contribute to alignment under any major theory of AI alignment (e.g. a pure capabilities project using the word "alignment" for some reason).

Get αΉ€600 play money
Sort by:

Does 10 billion dollars in RLHF compute count?

Compute dedicated specifically to RLHF/CAI/other compute-intensive alignment techniques, yes. Total compute allocated to a broader capabilities-focused project that happens to include some alignment elements, no.

More related questions