Either this market is not being very Bayesian, or people think I'm going to misresolve it, or it was always hella weighted towards the last couple of months of 2026.

Yeah, weird for the market price not to drift down as the end of 2026 approaches. I think people feel uncertain whether you think Hugging Face and Navier-Stokes and such already count as all hell breaking loose.

And to be explicit, I hear you as saying no, you do not count events so far as qualifying. Possible rule of thumb: if a non-technical, news-ignoring person can, with a straight face, dismiss AI as hype, then all hell hasn't broken loose. They might be being willfully blind, but if they can still do that, it's still potential as opposed to kinetic hell, as another commenter put it. I'm aiming for something similar in my "life-changing for muggles" market.

PS: Claude kindly points out that (putting it my own words) in theory the market price could rationally stay flat-ish if the surprisingness of new evidence like superhuman math keeps managing to roughly cancel the evidence from the clock running out on 2026.

Nervous about this one re: wave of open models around the end of the year. 40% for same question but "end of 2027"

Would "you from 10-15 years ago" be surprised at how low your anxiety levels are at this point? Considering the current level of capabilities.

"...all hell could break loose almost instantaneously."

https://youtube.com/shorts/nrlqJ-SzBVg?si=sMM9VaCwAQ_Hwcuy

@EliezerYudkowsky Does this market wait specifically until close to resolve, or can it resolve at any point in time? (Does a lack of a resolution imply that AI hell did not break loose yet?)

@Quroe Lack of a resolution means that I'm not certain that AI hell has already broken out.

What's the smallest, least hell break loose sequence of events you would nonetheless describe as all hell breaking loose?

Does this resolve based on global impact, or would a huge disruption to a single country be sufficient?

@EliezerYudkowsky What is your current estimate of the probability of this happening?

Which year or month (if any) of the AI 2027 race ending would you say is the earliest one such that it would count as loose-broken AI hell if the described events up to and including that period all transpired within 2026?

@TheAllMemeingEye Haven't read AI 2027 actually. :/ I asked Opus to summarize 2027 events by month and it sounds like "July: Agent-3-mini is released publicly, cheaper and better than a typical employee; OpenBrain declares AGI; massive job disruption, market frenzy, and public backlash ensue; external evaluators find Agent-3-mini can produce detailed bioweapon instructions" is when signs first become publicly visible so I'd go with July.