I'm pretty uncertain on this one. Could a superintelligence find a plan which fools me? Yes. Will such a plans show up early on in a search order without actively trying to fool me? Ehh... harder to say. It's definitely a possibility I keep in mind. Most importantly, over time as our understanding improves on the theory side, it gets less and less likely that a plan which would fool me shows up early in a natural search order.

Note that Eliezer and I and probably Tammy would all tell you that "can't see how it fails" is not a very useful bar to aim for. At the bare minimum, we should have both (1) a compelling positive case that it in fact works, and (2) a compelling positive case that even if we missed something crucial, failure is likely to be recoverable rather than catastrophic.

I don't know of one, though I haven't followed the literature much the past couple years.

Basically every time a new model is released by a major lab, I hear from at least one person (not always the same person) that it's a big step forward in programming capability/usefulness. And then David gives it a try, and it works qualitatively the same as everything else: great as a substitute for stack overflow, can do some transpilation if you don't mind generating kinda crap code and needing to do a bunch of bug fixes, and somewhere between useless and actively harmful on anything even remotely complicated.

It would be nice if there were someone who tries out every new model's coding capabilities shortly after they come out, reviews it, and gives reviews with a decent chance of actually matching David's or my experience using the thing (90% of which will be "not much change") rather than getting all excited every single damn time. But also, to be a useful signal, they still need to actually get excited when there's an actually significant change. Anybody know of such a source?

EDIT-TO-ADD: David has a comment below with a couple examples of coding tasks.

On my understanding, the push for centralization came from a specific faction whose pitch was basically:

• here's the scaling laws for tokamaks
• here's how much money we'd need
• ... so let's make one real big tokamak rather than spending money on lots of little research devices.

... and that faction mostly won the competition for government funding for about half a century.

The current boom accepted that faction's story at face value, but then noticed that new materials allowed the same "scale up the tokamaks" strategy to be executed on a budget achievable with private funding, and therefore they could fund projects without having to fight the faction which won the battle for government funding.

The counterfactual which I think is probably correct is that there exist entirely different designs far superior to tokamaks, which don't require that much scale in the first place, but which were never discovered because the "scale up the tokamaks" faction basically won the competition for funding and stopped most research on alternative designs from happening.

If you mean the Manhattan Project: no. IIUC there were basically zero Western groups and zero dollars working toward the bomb before that, so the Manhattan Project clearly sped things up. That's not really a case of "centralization" so much as doing-the-thing-at-all vs not-doing-the-thing-at-all.

If you mean fusion: yes. There were many fusion projects in the sixties, people were learning quickly. Then the field centralized, and progress slowed to a crawl.

I think this is missing the most important consideration: centralization would likely massively slow down capabilities progress.

the number one spontaneous conversation is "what are you working on" or "what have you done so far", which forces you to re-explain what you're doing & the reasons for doing it to a skeptical & ignorant audience

I'm very curious if others also find this to be the biggest value-contributor amongst spontaneous conversations. (Also, more generally, I'm curious what kinds of spontaneous conversations people are getting so much value out of.)

I have heard people say this so many times, and it is consistently the opposite of my experience. The random spontaneous conversations at conferences are disproportionately shallow and tend toward the same things which have been discussed to death online already, or toward the things which seem simple enough that everyone thinks they have something to say on the topic. When doing an activity with friends, it's usually the activity which is novel and/or interesting, while the conversation tends to be shallow and playful and fun but not as substantive as the activity. At work, spontaneous conversations generally had little relevance to the actual things we were/are working on (there are some exceptions, but they're rarely as high-value as ordinary work).

The ice cream snippets were good, but they felt too much like they were trying to be a relatively obvious not-very-controversial example of the problems you're pointing at, rather than a central/prototypical example. Which is good as an intro, but then I want to see it backed up by more central examples.

The dishes example was IMO the best in the post, more like that would be great.

Unfiltered criticism was discussed in the abstract, it wasn't really concrete enough to be an example. Walking through an example conversation (like the ice cream thing) would help.

Mono vs open vs poly would be a great example, but it needs an actual example conversation (like the ice cream thing), not just a brief mention. Same with career choice. I want to see how specifically the issues you're pointing to come up in those contexts.

(Also TBC it's an important post, and I'm glad you wrote it.)

This post would be a LOT better with about half-a-dozen representative real examples you've run into.

One example, to add a little concreteness: suppose that the path to AGI is to scale up o1-style inference-time compute, but it requires multiple OOMs of scaling. So it no longer has a relatively-short stream of "internal" thought, it's more like the natural-language record of an entire simulated society.

Then:

• There is no hope of a human reviewing the whole thing, or any significant fraction of the whole thing. Even spot checks don't help much, because it's all so context-dependent.
• Accurate summarization would itself be a big difficult research problem.
• There's likely some part of the simulated society explicitly thinking about intentional deception, even if the system as a whole is well aligned.
• ... but that's largely irrelevant, because in the context of a big complex system like a whole society, the effects of words are very decoupled from their content. Think of e.g. a charity which produces lots of internal discussion about reducing poverty, but frequently has effects entirely different from reducing poverty. The simulated society as a whole might be superintelligent, but its constituent simulated subagents are still pretty stupid (like humans), so their words decouple from effects (like humans' words).

... and that's how the proposal breaks down, for this example.

I haven't decided yet whether to write up a proper "Why Not Just..." for the post's proposal, but here's an overcompressed summary. (Note that I'm intentionally playing devil's advocate here, not giving an all-things-considered reflectively-endorsed take, but the object-level part of my reflectively-endorsed take would be pretty close to this.)

Charlie's concern isn't the only thing it doesn't handle. The only thing this proposal does handle is an AI extremely similar to today's, thinking very explicitly about intentional deception, and even then the proposal only detects it (as opposed to e.g. providing a way to solve the problem, or even a way to safely iterate without selecting against detectability). And that's an extremely narrow chunk of the X-risk probability mass - any significant variation in the AI breaks it, any significant variation in the threat model breaks it. The proposal does not generalize to anything.

Charlie's concern is just one specific example of a way in which the proposal does not generalize. A proper "Why Not Just..." post would list a bunch more such examples.

And as with Charlie's concern, the meta-level problem is that the proposal also probably wouldn't get us any closer to handling those more-general situations. Sure, we could make some very toy setups (like the chess thing), and see what the shoggoth+face AI does on those very toy setups, but we get very few bits, and the connection is very tenuous to both other threat models and AIs with any significant differences from the shoggoth+face. Accounting for the inevitable failure to measure what we think we're measuring (with probability close to 1), such experiments would not actually get us any closer to solving any of the problems which constitute the bulk of the X-risk probability mass. It's not "a start", because "a start" would imply that the experiment gets us closer, i.e. that the problem gets easier after doing the experiment. If you try to think about the You Are Not Measuring What You Think You Are Measuring problem as "well, we got at least some tiny epsilon of evidence, right?", then you will shoot yourself in the foot; such reasoning is technically correct, but the correct value of epsilon is small enough that the correct update from it is not distinguishable from zero in practice.

The problem with that sort of attitude is that, when the "experiment" yields so few bits and has such a tenuous connection to the thing we actually care about (as in Charlie's concern), that's exactly when You Are Not Measuring What You Think You Are Measuring bites real hard. Like, sure, you'll see this system do something in the toy chess experiment, but that's just not going to be particularly relevant to the things an actual smarter-than-human AI does in the situations Charlie's concerned about. If anything, the experimenter is far more to likely to fool themselves into thinking their results are relevant to Charlie's concern than they are to correctly learn anything relevant to Charlie's concern.

I think this misunderstands what discussion of "barriers to continued scaling" is all about. The question is whether we'll continue to see ROI comparable to recent years by continuing to do the same things. If not, well... there is always, at all times, the possibility that we will figure out some new and different thing to do which will keep capabilities going. Many people have many hypotheses about what those new and different things could be: your guess about interaction is one, inference time compute is another, synthetic data is a third, deeply integrated multimodality is a fourth, and the list goes on. But these are all hypotheses which may or may not pan out, not already-proven strategies, which makes them a very different topic of discussion than the "barriers to continued scaling" of the things which people have already been doing.

Some of the underlying evidence, like e.g. Altman's public statements, is relevant to other forms of scaling. Some of the underlying evidence, like e.g. the data wall, is not. That cashes out to differing levels of confidence in different versions of the prediction.

Oh I see, you mean that the observation is weak evidence for the median model relative to a model in which the most competent researchers mostly determine memeticity, because higher median usually means higher tails. I think you're right, good catch.

FYI, my update from this comment was:

• Hmm, seems like a decent argument...
• ... except he said "we don't know that it doesn't work", which is an extremely strong update that it will clearly not work.

Still very plausible as a route to continued capabilities progress. Such things will have very different curves and economics, though, compared to the previous era of scaling.

I don't expect that to be particularly relevant. The data wall is still there; scaling just compute has considerably worse returns than the curves we've been on for the past few years, and we're not expecting synthetic data to be anywhere near sufficient to bring us close to the old curves.

unless you additionally posit an additional mechanism like fields with terrible replication rates have a higher standard deviation than fields without them

Why would that be relevant?

Regarding the recent memes about the end of LLM scaling: David and I have been planning on this as our median world since about six months ago. The data wall has been a known issue for a while now, updates from the major labs since GPT-4 already showed relatively unimpressive qualitative improvements by our judgement, and attempts to read the tea leaves of Sam Altman's public statements pointed in the same direction too. I've also talked to others (who were not LLM capability skeptics in general) who had independently noticed the same thing and come to similar conclusions.

Our guess at that time was that LLM scaling was already hitting a wall, and this would most likely start to be obvious to the rest of the world around roughly December of 2024, when the expected GPT-5 either fell short of expectations or wasn't released at all. Then, our median guess was that a lot of the hype would collapse, and a lot of the investment with it. That said, since somewhere between 25%-50% of progress has been algorithmic all along, it wouldn't be that much of a slowdown to capabilities progress, even if the memetic environment made it seem pretty salient. In the happiest case a lot of researchers would move on to other things, but that's an optimistic take, not a median world.

(To be clear, I don't think you should be giving us much prediction-credit for that, since we didn't talk about it publicly. I'm posting mostly because I've seen a decent number of people for whom the death of scaling seems to be a complete surprise and they're not sure whether to believe it. For those people: it's not a complete surprise, this has been quietly broadcast for a while now.)

I am posting this now mostly because I've heard it from multiple sources. I don't know to what extent those sources are themselves correlated (i.e. whether or not the rumor started from one person).

Epistemic status: rumor.

Word through the grapevine, for those who haven't heard: apparently a few months back OpenPhil pulled funding for all AI safety lobbying orgs with any political right-wing ties. They didn't just stop funding explicitly right-wing orgs, they stopped funding explicitly bipartisan orgs.

I don't think statistics incompetence is the One Main Thing, it's just an example which I expect to be relatively obvious and legible to readers here.

The way I think of it, it's not quite that some abstractions are cheaper to use than others, but rather:

• One can in-principle reason at the "low(er) level", i.e. just not use any given abstraction. That reasoning is correct but costly.
• One can also just be wrong, e.g. use an abstraction which doesn't actually match the world and/or one's own lower level model. Then predictions will be wrong, actions will be suboptimal, etc.
• Reasoning which is both cheap and correct routes through natural abstractions. There's some degrees of freedom insofar as a given system could use some natural abstractions but not others, or be wrong about some things but not others.

Proof that the quoted bookkeeping rule works, for the exact case:

• The original DAG $G$ asserts $P\left[X\right]={\prod }_{i}P\left[X_i|X_{p{a}^G\left(i\right)}\right]$
• If $G^{\prime }$ just adds an edge from $j$ to $k$, then $G^{\prime }$ says $P\left[X\right]=P[X_k|X_{p{a}^G\left(k\right)},X_j]{\prod }_{i\ne k}P\left[X_i|X_{p{a}^G\left(i\right)}\right]$
• The original DAG's assertion $P\left[X\right]={\prod }_{i}P\left[X_i|X_{p{a}^G\left(i\right)}\right]$ also implies $P[X_k|X_{p{a}^G\left(k\right)},X_j]=P\left[X_k|X_{p{a}^G\left(k\right)}\right]$, and therefore implies $G^{\prime }$'s assertion $P\left[X\right]=P[X_k|X_{p{a}^G\left(k\right)},X_j]{\prod }_{i\ne k}P\left[X_i|X_{p{a}^G\left(i\right)}\right]$.

The approximate case then follows by the new-and-improved Bookkeeping Theorem.

Not sure where the disconnect/confusion is.

All dead-on up until this:

... the universe will force them to use the natural abstractions (or else fail to achieve their goals). [...] Would the argument be that unnatural abstractions are just in practice not useful, or is it that the universe is such that its ~impossible to model the world using unnatural abstractions?

It's not quite that it's impossible to model the world without the use of natural abstractions. Rather, it's far instrumentally "cheaper" to use the natural abstractions (in some sense). Rather than routing through natural abstractions, a system with a highly capable world model could instead e.g. use exponentially large amounts of compute (e.g. doing full quantum-level simulation), or might need enormous amounts of data (e.g. exponentially many training cycles), or both. So we expect to see basically-all highly capable systems use natural abstractions in practice.

The problem with this model is that the "bad" models/theories in replication-crisis-prone fields don't look like random samples from a wide posterior. They have systematic, noticeable, and wrong (therefore not just coming from the data) patterns to them - especially patterns which make them more memetically fit, like e.g. fitting a popular political narrative. A model which just says that such fields are sampling from a noisy posterior fails to account for the predictable "direction" of the error which we see in practice.

Walking through your first four sections (out of order):

• Systems definitely need to be interacting with mostly-the-same environment in order for convergence to kick in. Insofar as systems are selected on different environments and end up using different abstractions as a result, that doesn't say much about NAH.
• Systems do not need to have similar observational apparatus, but the more different the observational apparatus the more I'd expect that convergence requires relatively-high capabilities. For instance: humans can't see infrared/UV/microwave/radio, but as human capabilities increased all of those became useful abstractions for us.
• Systems do not need to be subject to similar selection pressures/constraints or have similar utility functions; a lack of convergence among different pressures/constraints/utility is one of the most canonical things which would falsify NAH. That said, the pressures/constraints/utility (along with the environment) do need to incentivize fairly general-purpose capabilities, and the system needs to actually achieve those capabilities.

More general comment: the NAH says that there's a specific, discrete set of abstractions in any given environment which are "natural" for agents interacting with that environment. The reason that "general-purpose capabilities" are relevant in the above is that full generality and capability requires being able to use ~all those natural abstractions (possibly picking them up on the fly, sometimes). But a narrower or less-capable agent will still typically use some subset of those natural abstractions, and factors like e.g. similar observational apparatus or similar pressures/utility will tend to push for more similar subsets among weaker agents. Even in that regime, nontrivial NAH predictions come from the discreteness of the set of natural abstractions; we don't expect to find agents e.g. using a continuum of abstractions.

Our broader society has community norms which require basically everyone to be literate. Nonetheless, there are jobs in which one can get away without reading, and the inability to read does not make it that much harder to make plenty of money and become well-respected. These statements are not incompatible.

since for some reason using "memetic" in the same way feels very weird or confused to me, like I would almost never say "this has memetic origin"

... though now that it's been pointed out, I do feel like I want a short handle for "this idea is mostly passed from person-to-person, as opposed to e.g. being rederived or learned firsthand".

I also kinda now wish "highly genetic" meant that a gene has high fitness, that usage feels like it would be more natural.

I have split feelings on this one. On the one hand, you are clearly correct that it's useful to distinguish those two things and that my usage here disagrees with the analogous usage in genetics. On the other hand, I have the vague impression that my usage here is already somewhat standard, so changing to match genetics would potentially be confusing in its own right.

It would be useful to hear from others whether they think my usage in this post is already standard (beyond just me), or they had to infer it from the context of the post. If it's mostly the latter, then I'm pretty sold on changing my usage to match genetics.

It is indeed an anology to 'genetic'. Ideas "reproduce" via people sharing them. Some ideas are shared more often, by more people, than others. So, much like biologists think about the relative rate at which genes reproduce as "genetic fitness", we can think of the relative rate at which ideas reproduce as "memetic fitness". (The term comes from Dawkins back in the 70's; this is where the word "meme" originally came from, as in "internet memes".)

I hadn't made that connection yet. Excellent insight, thank you!

Y'know, you probably have the data to do a quick-and-dirty check here. Take a look at the GRE/SAT scores on the applications (both for applicant pool and for accepted scholars). If most scholars have much-less-than-perfect scores, then you're probably not hiring the top tier (standardized tests have a notoriously low ceiling). And assuming most scholars aren't hitting the test ceiling, you can also test the hypothesis about different domains by looking at the test score distributions for scholars in the different areas.

I suspect that your estimation of "how smart do these people seem" might be somewhat contingent on research taste. Most MATS research projects are in prosaic AI safety fields like oversight & control, evals, and non-"science of DL" interpretability, while most PIBBSS research has been in "biology/physics-inspired" interpretability, agent foundations, and (recently) novel policy approaches (all of which MATS has supported historically).

I think this is less a matter of my particular taste, and more a matter of selection pressures producing genuinely different skill levels between different research areas. People notoriously focus on oversight/control/evals/specific interp over foundations/generalizable interp because the former are easier. So when one talks to people in those different areas, there's a very noticeable tendency for the foundations/generalizable interp people to be noticeably smarter, more experienced, and/or more competent. And in the other direction, stronger people tend to be more often drawn to the more challenging problems of foundations or generalizable interp.

So possibly a MATS apologist reply would be: yeah, the MATS portfolio is more loaded on the sort of work that's accessible to relatively-mid researchers, so naturally MATS ends up with more relatively-mid researchers. Which is not necessarily a bad thing.

Fellows.

No, I meant that the correlation between pay and how-competent-the-typical-participant-seems-to-me is, if anything, negative. Like, the hiring bar for Google interns is lower than any of the technical programs, and PIBBSS seems-to-me to have the most competent participants overall (though I'm not familiar with some of the programs).

My main metric is "How smart do these people seem when I talk to them or watch their presentations?". I think they also tend to be older and have more research experience.

To be clear, I don't really think of myself as libertarian these days, though I guess it'd probably look that way if you just gave me a political alignment quiz.

To answer your question: I'm two years older than my brother, who is two years older than my sister.

Y'know @Ryan, MATS should try to hire the PIBBSS folks to help with recruiting. IMO they tend to have the strongest participants of the programs on this chart which I'm familiar with (though high variance).

... WOW that is not an efficient market. 

My two siblings and I always used to trade candy after Halloween and Easter. We'd each lay out our candy on table, like a little booth, and then haggle a lot.

My memories are fuzzy, but apparently the way this most often went was that I tended to prioritize quantity moreso than my siblings, wanting to make sure that I had a stock of good candy which would last a while. So naturally, a few days later, my siblings had consumed their tastiest treats and complained that I had all the best candy. My mother then stepped in and redistributed the candy.

And that was how I became a libertarian at a very young age.

Only years later did we find out that my mother would also steal the best-looking candies after we went to bed and try to get us to blame each other, which... is altogether too on-the-nose for this particular analogy.

Conjecture's Compendium is now up. It's intended to be a relatively-complete intro to AI risk for nontechnical people who have ~zero background in the subject. I basically endorse the whole thing, and I think it's probably the best first source to link e.g. policymakers to right now.

I might say more about it later, but for now just want to say that I think this should be the go-to source for new nontechnical people right now.

I think that conclusion is basically correct.

Part of the problem is that the very large majority of people I run into have minds which fall into a relatively low-dimensional set and can be "ray traced" with fairly little effort. It's especially bad in EA circles.

Human psychology, mainly. "Dominance"-in-the-human-intuitive-sense was in the original post mainly because I think that's how most humans intuitively understand "power", despite (I claimed) not being particularly natural for more-powerful agents. So I'd expect humans to be confused insofar as they try to apply those dominance-in-the-human-intuitive-sense intuitions to more powerful agents.

And like, sure, one could use a notion of "dominance" which is general enough to encompass all forms of conflict, but at that point we can just talk about "conflict" and the like without the word "dominance"; using the word "dominance" for that is unnecessarily confusing, because most humans' intuitive notion of "dominance" is narrower.

Because there'd be an unexploitable-equillibrium condition where a government that isn't focused on dominance is weaker than a government more focused on government, it would generally be held by those who have the strongest focus on dominance.

This argument only works insofar as governments less focused on dominance are, in fact, weaker militarily, which seems basically-false in practice in the long run. For instance, autocratic regimes just can't compete industrially with a market economy like e.g. most Western states today, and that industrial difference turns into a comprehensive military advantage with relatively moderate time and investment. And when countries switch to full autocracy, there's sometimes a short-term military buildup but they tend to end up waaaay behind militarily a few years down the road IIUC.

A monopoly on violence is not the only way to coordinate such things - even among humans, at small-to-medium scale we often rely on norms and reputation rather than an explicit enforcer with a monopoly on violence. The reason those mechanisms don't scale well for humans seems to be (at least in part) that human cognition is tuned for Dunbar's number.

And even if a monopoly on violence does turn out to be (part of) the convergent way to coordinate, that's not-at-all synonymous with a dominance hierarchy. For instance, one could imagine the prototypical libertarian paradise in which a government with monopoly on violence enforces property rights and contracts but otherwise leaves people to interact as they please. In that world, there's one layer of dominance, but no further hierarchy beneath. That one layer is a useful foundational tool for coordination, but most of the day-to-day work of coordinating can then happen via other mechanisms (like e.g. markets).

(I suspect that a government which was just greedy for resources would converge to an arrangement roughly like that, with moderate taxes on property and/or contracts. The reason we don't see that happen in our world is mostly, I claim, that the humans who run governments usually aren't just greedy for resources, and instead have a strong craving for dominance as an approximately-terminal goal.)