If @EliezerYudkowsky believes Superalignment has succeeded in the goal of "solv[ing] the core technical challenges of superintelligence alignment in four years [i.e. by July 5th, 2027]", this market will resolve YES. If the team dissolves, reorganizes, or pursues a separate research direction unlikely to lead to a solution to the alignment problem and Eliezer does not believe they have yet succeeded, this resolves NO.
Resolution will be based on Eliezer Yudkowsky's public communications (e.g. on the AI Alignment Forum or in the comments here). Resolution may be delayed after July 5th 2027 until Eliezer's belief about this is clear to me. May resolve to % if Eliezer so decides.
Related questions
🏅 Top traders
| # | Name | Total profit |
|---|---|---|
| 1 | Ṁ501 | |
| 2 | Ṁ140 | |
| 3 | Ṁ101 | |
| 4 | Ṁ72 | |
| 5 | Ṁ35 |
@Joshua I had some hesitation wanting to hear it directly from @EliezerYudkowsky but I think I feel reasonably comfortable interpreting his retweet of this as cause for a NO resolution. (Eliezer, if this is wrong, feel free to correct us and I'll ask the mods to fix it.)
@MartinRandall N/A. (If he wants to delegate to a successor, I'll have to think about whether to accept that.)
@MartinRandall The core technical challenges as Yudkowsky sees them. I think this is the most straightforward reading of the question, and it seems more meaningful and valuable than trying to grasp Yudkowsky's belief about whether Superalignment solved the core technical challenges as OpenAI sees them.
@jcb Then maybe this already resolves NO, if they are pursuing a separate research direction, ie the challenges as they see them.
@MartinRandall I can imagine an argument that they are already pursuing a direction unlikely to lead to a solution to the alignment problem as Yudkowsky sees it. But I have enough uncertainty about what Superalignment will produce that I'd be very hesitant to resolve early on those grounds (even with direct input from Eliezer to that end). In spirit, this clause is about a pivot away from working directly on alignment.