[HN Gopher] A deep critique of AI 2027's bad timeline models
       ___________________________________________________________________
        
       A deep critique of AI 2027's bad timeline models
        
       Author : paulpauper
       Score  : 67 points
       Date   : 2025-06-23 18:51 UTC (4 hours ago)
        
 (HTM) web link (www.lesswrong.com)
 (TXT) w3m dump (www.lesswrong.com)
        
       | brcmthrowaway wrote:
       | So much bikeshedding and armchair expertise displayed in this
       | field.
        
         | goatlover wrote:
         | Which would that be, the arguments for ASI being near and how
         | that could be apocalyptic, or the push back on those timelines
         | and doomsday (or utopian) proclamations?
        
           | evilantnie wrote:
           | I don't think the real divide is "doom tomorrow" vs "nothing
           | to worry about." The crux is a pretty straightforward
           | philosophical question "what does it even mean to generalize
           | intelligence and agency", how much can scaling laws tell us
           | about that?
           | 
           | The back-and-forth over s2's and growth exponents feels like
           | theatrics that bury the actual debate.
        
             | vonneumannstan wrote:
             | >The crux is a pretty straightforward philosophical
             | question "what does it even mean to generalize intelligence
             | and agency", how much can scaling laws tell us about that?
             | 
             | Truly a bizarre take. I'm sure the Dinosaurs also debated
             | the possible smell and taste of the asteroid that was about
             | to hit them. The real debate. lol.
        
               | evilantnie wrote:
               | The dinosaurs didn't create the asteroid that hit them,
               | so they never had the chance for a real debate.
        
         | ysofunny wrote:
         | with all that signaling... it's almost like they're trying to
         | communicate!!! who would've thought!?
        
       | TimPC wrote:
       | This critique is fairly strong and offers a lot of insight into
       | the critical thinking behind it. The parts of the math I've
       | looked at do check out.
        
       | yodon wrote:
       | So... both authors predict superhuman intelligence, defined as AI
       | that can complete tasks that would take humans hundreds of hours,
       | to be a thing "sometime in the next few years", both authors
       | predict "probably not before 2027, but maybe" and both authors
       | predict "probably not longer than 2032, but maybe", and one
       | author seems to think their estimates are wildly better than
       | those of the other author.
       | 
       | That's not quite the level of disagreement I was expecting given
       | the title.
        
         | jollyllama wrote:
         | That's not very investor of you
        
         | LegionMammal978 wrote:
         | As far as I can tell, the author of the critique specifically
         | avoids espousing a timeline of his own. Indeed, he dislikes how
         | these sorts of timeline models are used in general:
         | 
         | > I'm not against people making shoddy toy models, and I think
         | they can be a useful intellectual exercise. I'm not against
         | people sketching out hypothetical sci-fi short stories, I've
         | done that myself. I am against people treating shoddy toy
         | models as rigorous research, stapling them to hypothetical
         | short stories, and then taking them out on podcast circuits to
         | go viral. What I'm most against is people taking shoddy toy
         | models seriously and _basing life decisions on them_ , as I
         | have seen happen for AI2027. This is just a model for a tiny
         | slice of the possibility space for how AI will go, and in my
         | opinion it is implemented poorly even if you agree with the
         | author's general worldview.
         | 
         | In particular, I wouldn't describe the author's position as
         | "probably not longer than 2032" (give or take the usual
         | quibbles over what tasks are a necessary part of "superhuman
         | intelligence"). Indeed, he rates social issues from AI as a
         | more plausible near-term threat than dangerous AGI takeoff [0],
         | and he is very skeptical about how well any software-based AI
         | can revolutionize the physical sciences [1].
         | 
         | [0] https://titotal.substack.com/p/slopworld-2035-the-dangers-
         | of...
         | 
         | [1] https://titotal.substack.com/p/ai-is-not-taking-over-
         | materia...
        
           | ysofunny wrote:
           | but what is the difference between a shoddy toy model and a
           | real world pro "rigorous research"?
           | 
           | it's like asking between the difference between amateur toy
           | audio gear, and real pro level audio gear... (which is not a
           | simple thing given "prosumer products" dominate the
           | landscape)
           | 
           | the only point in betting when "real AGI" will happen boils
           | down to the payouts from gambling with this. are such gambles
           | a zero sum game? does that depend on who escrows the bet??
           | 
           | what do I get if I am correct? how should the incorrect lose?
        
             | LegionMammal978 wrote:
             | If you believe that there's any plausible chance of AGI
             | causing a major catastrophe short of the immediate end of
             | the world, then its precise nature can have all sorts of
             | effects on how the catastrophe could unfold and how people
             | should respond to it.
        
         | vonneumannstan wrote:
         | For rationalists this is about as bad as disagreements can
         | get...
        
         | TimPC wrote:
         | He predicts it might be possible from model math but doesn't
         | actually say what his prediction is. He also argues it's
         | possible we are on a s-curve that levels out before superhuman
         | intelligence.
        
         | sweezyjeezy wrote:
         | I don't think the author of this article is making any strong
         | prediction, in fact I think a lot of the article is a critique
         | of whether such an extrapolation can be done meaningfully.
         | 
         |  _Most of these models predict superhuman coders in the near
         | term, within the next ten years. This is because most of them
         | share the assumption that a) current trends will continue for
         | the foreseeable future, b) that "superhuman coding" is possible
         | to achieve in the near future, and c) that the METR time
         | horizons are a reasonable metric for AI progress. I don't agree
         | with all these assumptions, but I understand why people that do
         | think superhuman coders are coming soon._
         | 
         | Personally I think any model that puts zero weight on the idea
         | that there could be some big stumbling blocks ahead, or even a
         | possible plateau, is not a good model.
        
           | XorNot wrote:
           | The primary question is always whether they'd have made those
           | sorts of predictions based on the results they were seeing on
           | the field from the same amount of time in the past.
           | 
           | Pre-CharGPT I very much doubt the bullish predictions on AI
           | would've been made the way they are now.
        
       | lubujackson wrote:
       | These predictions seem wildly reductive in any case and it seems
       | like extrapolating AI's ability to complete task that would take
       | a human 30 seconds -> 10 minutes is far different than going from
       | 10 minutes to 5 years. For one reason, a 5 year task generally
       | requires much more input and intent than a 10 minute task.
       | Already we have ramped up from "enter a paragraph" to complicated
       | Cursor rules and rich context prompts to get to where we are
       | today. This is completely overlooked in these simple "graphs go
       | up" predictions.
        
         | echelon wrote:
         | I'm also interested in error rates multiplying for simple
         | tasks.
         | 
         | A human can do a long sequence of easy tasks without error - or
         | easily correct. Can a model do the same?
        
           | kingstnap wrote:
           | The recent Apple "LLMs can't reason yet" paper was exactly
           | this. They just tested if models could run an exponential
           | number of steps.
           | 
           | Of course, they gave it a terrible clickbait title and framed
           | the question and graphs incorrectly. But if they did the
           | study better it would have been "How long of a sequence of
           | algorithmic steps can LLMs execute before making a mistake or
           | giving up?"
        
       | kypro wrote:
       | As someone in the P(doom) > 90% category, I think in general
       | making overly precise predictions are a really bad way to
       | highlight AI risks (assuming that was the goal of AI 2027).
       | 
       | Making predictions that are too specific just opens you up to
       | pushback from people who are more interested in critiquing the
       | exact details of your softer predictions (such as those around
       | timelines) rather than your hard predictions about likely
       | outcomes. And while I think articles like this are valuable to
       | refine timeline predictions, I find a lot of people use them as
       | evidence to dismiss the stronger predictions made about the risks
       | of ASI.
       | 
       | I think people like Nick Bostrom make much more convincing
       | arguments about AI risk because they don't depend on overly
       | detailed predictions which can be easily nit-picked at, but are
       | instead much more general and focus more on the unique nature of
       | the risks AI presents.
       | 
       | For me the risk of timelines is that they're unknowable due to
       | the unpredictable nature of ASI. The fact we are rapidly
       | developing a technology which most people would accept comes with
       | at least some existential risk, that we can't predict the
       | progress curve of, and where solutions would come with
       | significant coordination problems should concern people without
       | having to say it will happen in x number of years.
       | 
       | I think AI 2027 is interesting as a science fiction about
       | potential futures we could be heading towards, but that's really
       | it.
       | 
       | The problem with being an AI doomer is that you can't say "I told
       | you so" if you're right so any personal predictions you make have
       | no close to no expected pay-out, either socially or economically.
       | This is different to other risks which if you predict accurately
       | when others don't you can still benefit from.
       | 
       | I have no meaningful voice in this space so I'll just keep saying
       | we're fucked because what does it matter what I think, but I wish
       | there were more people with influence out there who were
       | seriously thinking about how they can best influence rather than
       | stroking their own own egos with future predictions, which even
       | if I happen agree with do next to nothing to improve the
       | distribution of outcomes.
        
         | Fraterkes wrote:
         | I'm not trying to be disingenuous, but in what ways have you
         | changed your life now that you belief theres >90% chance to an
         | end to civilization/humanity? Are you living like a terminal
         | cancer patient?
         | 
         | (Im sorry, I know its a crass question)
        
           | allturtles wrote:
           | The person you're replying to said "For me the risk of
           | timelines is that they're unknowable due to the unpredictable
           | nature of ASI." So they are predicting >90% chance of doom,
           | but not when that will happen. Given that there is already a
           | 100% chance of death at some unknown point in the future, why
           | would this cause GP to start living like a terminal cancer
           | patient (presumably defined as someone with a >99% chance of
           | death in the next year)?
        
             | lava_pidgeon wrote:
             | I like to point out, that the existence of AGI in the
             | future does change my potential future planning. So I am
             | 35. Do I need save for pensions? Does it make to sense to
             | start family? These aren't 1 year questions but 20 years
             | ahead questions...
        
               | amarcheschi wrote:
               | If you're so terrorized of Ai to not start a family
               | despite wanting and being able to, it must be miserable
               | (if) to eventually live through the years as everyone
               | that tried to predict the end of the world did (except
               | for those who died of other causes before the predicted
               | end)
        
         | siddboots wrote:
         | I think both approaches are useful. AI2027 presents a specific
         | timeline in which a) the trajectory of tech is at least
         | somewhat empirically grounded, and b) each step of the plot arc
         | is plausible. There's a chance of it being convincing to a
         | skeptic who had otherwise thought of the whole "rogue AI"
         | scenario as a kind of magical thinking.
        
           | kypro wrote:
           | I agree, but I think you're assuming a certain type of person
           | who understands that a detailed prediction can be both wrong
           | and right simultaneously. And that it's not so much about
           | getting all the details right, but being in the right
           | ballpark.
           | 
           | Unfortunately there's a huge number of people who get
           | obsessed about details and then nit pick. I see this with
           | Eliezer Yudkowsky all the time where 90% of the criticism of
           | his views are just nit picking the weaker predictions he
           | makes while ignoring his stronger predictions regarding the
           | core risks which could result in those bad things happening.
           | I think Yudkowsky opens himself up to this though because he
           | often makes very detailed predictions about how things might
           | play out and this largely why he's so controversial, in my
           | opinion.
           | 
           | I really liked AI 2027 personally. I thought specifically the
           | tabletop exercises were a nice heuristic for predicting how
           | actors might behave in certain scenarios. I also agree that
           | it presented a plausible narrative for how things could play
           | out. I'm also glad they did wimp out with the bad ending.
           | Another problem I have with people are concerned about AI
           | risk is that they scare away from speaking plainly about the
           | fact if things go poorly your love ones in a few years will
           | probably either be either be dead, in suspended animation on
           | a memory chip, or in a literal digital hell.
        
       | boznz wrote:
       | I expect the predictions for fusion back in the 1950's and 1960's
       | generated similar essays, they had not got to ignition but the
       | science was solid; the 'science' with moving from AGI to ASI is
       | not really that solid yet we have yet to achieve 'AI ignition'
       | even in the lab. (Any AI's that have achieved consciousness feel
       | free to disagree)
        
         | fasthands9 wrote:
         | I do agree generally with this, but AI 2027 and other writings
         | have moved my concern from 0% to 10%.
         | 
         | I know I sound crazy writing it out, but many of the really bad
         | scenarios don't require consciousness or anything like that. It
         | just requires they be self-replicating and the ability to
         | operate without humans shutting them off.
        
       | staunton wrote:
       | This is a lot of text, details and hair splitting just to say
       | "modeling things like this is bullshit". It's engaging
       | "seriously" and "on the merits" with something that from the very
       | start was just marketing fluff packaged as some kind of
       | prediction.
       | 
       | I'm not sure if the author did anyone a favor with this write-up.
       | More than anything, it buries the main point ("this kind of
       | forecasting is fundamentally bullshit") under a bunch of
       | complicated-sounding details that lend credibility to the
       | original predictions, which the original authors now get to agrue
       | about and thank people for pointing out "minor issues which we
       | have now addressed in the updated version".
        
       | ed wrote:
       | Anyone old enough to remember EPIC 2014? It was a viral flash
       | video, released in 2004, about the future of Google and news
       | reporting. I imagine 2027 will age similarly well.
       | 
       | https://youtu.be/LZXwdRBxZ0U
        
       | mlsu wrote:
       | My niece weighed 3 kg one year ago. Now, she weighs 8.9 kg. By my
       | modeling, she will weigh more than the moon in approximately 50
       | years. I've analyzed the errors in my model; regrettably, the
       | conclusion is always the same: it will certainly happen within
       | our lifetimes.
       | 
       | Everyone needs to be planning for this -- all of this urgent talk
       | of "AI" (let alone "climate change" or "holocene extinction") is
       | of positively no consequence compared to the prospect I've
       | outlined here: a mass of HUMAN FLESH the size of THE MOON growing
       | on the surface of our planet!
        
         | habinero wrote:
         | LOL, exactly. All of the weird AGI/doomer/whatever bullshit
         | we're calling it/ feels like exactly this: people who think
         | they're too smart to fall prey to groupthink and bias
         | confirmation, and yet predictably are falling prey to
         | groupthink and bias confirmation.
        
           | mlsu wrote:
           | I have more fun reading it as a kind of collaborative real-
           | time sci-fi story. Reads right out of a Lem novel.
        
         | Workaccount2 wrote:
         | We have watched many humans grow so we have a pretty good idea
         | of the curve. A better analogy is an alien blob appeared one
         | day and went from 3kg to 9kg in a year. We have never seen one
         | of these before, so we don't know what it's growth curve looks
         | like. But it keeps eating food and keeps getting bigger.
        
           | mlsu wrote:
           | Mine's different. She's cuter.
           | 
           | On a more serious note. Have these AI doom guys ever dealt
           | with one of these cutting edge models on _out of distribution
           | data_? They suck so so bad. There 's only so much data
           | available, the models have basically slurped it all.
           | 
           | Let alone like the basic thermodynamics of it. There's only
           | so much entropy out there in cyberspace to harvest, at some
           | point you run into a wall and then you have to build real
           | robots to go collect more in the real world. And how's that
           | going for them?
           | 
           | Also I can't help remarking: the metaphor you chose is
           | _science fiction_.
        
       ___________________________________________________________________
       (page generated 2025-06-23 23:01 UTC)