Max H

Karma: 3,395

Most of my posts and comments are about AI and alignment. Posts I’m most proud of, which also provide a good introduction to my worldview:

I also created Forum Karma, and wrote a longer self-introduction here.

PMs and private feedback are always welcome.

NOTE: I am not Max Harms, author of Crystal Society. I’d prefer for now that my LW postings not be attached to my full name when people Google me for other reasons, but you can PM me here or on Discord (m4xed) if you want to know who I am.

Max H 13 Jun 2026 3:03 UTC
29 points
16
in reply to: Philipreal’s comment on: Philipreal’s Shortform
We have reviewed a report that we believe is the basis of the government’s directive and validated that the level of capability displayed there is widely available from other models (including OpenAI’s GPT-5.5), and is used every day by the defenders who keep systems safe.
...
If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.
Export controls on AI are probably inevitable (and perhaps a good thing if done well), but this is pretty clearly retaliatory / punitive. OpenAI will presumably release a comparable model in the next few weeks or months, it will be similarly jailbreak-able, but not subject to the same restrictions. Then what?

Max H 4 Jun 2026 16:52 UTC
2 points
−2
in reply to: Mo Putera’s comment on: Max H’s Shortform
Yes, a few examples / reasons to leave off evidence and avoid (over)-explanation (spoilers):
1. Pedagogy. In any good math or physics text, many problems are “left as an exercise to the reader”. There’s a generalization of this principle that I think applies to not-strictly-pedagological discourse; e.g. nudging your interlocutor to generate your side of an argument and fill in the details on the fly makes them more likely to understand it more deeply and correctly.
2. Balancing out an effort asymmetry in debate / discourse. e.g. suppose someone makes a contentious top-level post, with some amount of evidence / support. Many readers will hold off on forming an opinion / judgement / vote until another party has had a chance to respond, and will then form their opinion based on the quality of the response(s). This creates an obligation to respond, which is a negative externality on the discourse participants, but also bad for the epistemic hygiene of the viewers. It would often be better for viewers to do the work of anticipating the best possible rebuttals they expect to see (and counter-rebuttals), and then make up their mind independently, without actually needing all the discourse to be generated for them.
3. Increasing engagement. A brief, provocative claim will often get more attention and invite more discussion than a longpost with all of the evidence, explanation, and anticipated responses to objections pre-written. This is not necessarily always positive / pro-social (engagement baiting), but is good if you think that increased discourse on your chosen topic is good.

Max H 4 Jun 2026 14:16 UTC
11 points
0
on: Max H’s Shortform
There’s a norm / explicit assumption that it’s always virtuous to support your claims with evidence and explanation. But sometimes it is actually better to say (what you believe to be) true things without explanation, and let readers do the work of generating evidence and support themselves.

Max H 10 May 2026 16:23 UTC
7 points
1
on: International Law Cannot Prevent Extinction Either
Treaties are real to the degree that they are backed up by credible threat of enforcement + deterrence for non-compliance, in the same way that ordinary national and local law is real to the degree that it is actually enforced. There are lots of laws on the books in many places that are not actually enforced consistently or at all, for various reasons, some of which are analogous to the issues with international treaties that you gesture at. And this in fact causes all sorts of problems in many cases—disorder, unfairness / injustice, degrading trust and legitimacy of the legal system and the state, etc. But it doesn’t follow that the entire legal system of any given nation is largely fiction, nor that we should stop trying to pass new laws or enforce existing ones, and it definitely doesn’t mean that we should resort to some other kind of system where the state does not have a monopoly on violence. Analogously, it would be premature and fatalistic to give up on international treaties for AGI (and it’s not clear what the alternative could possibly be—if the leaders of superpowers are not the ones ultimately deciding what to do about AGI, who is?)
Also:
I agree with Eliezer’s main thesis that individual violence against AI researchers is both morally wrong and strategically stupid. Where I disagree is with the claim that international law can prevent extinction.
“International law” is a vague term that means different things to different people; the linked post you’re criticizing doesn’t use that phrase at all (and is mostly not about the topic), and you don’t say yourself exactly what you mean by it except via example.

Max H 8 May 2026 6:12 UTC
8 points
6
in reply to: leogao’s comment on: leogao’s Shortform
higher margins rightfully means higher market cap. if your company is barely scraping by, youre not producing as much value.
Right, I think the market caps are justified for the most part. But market caps represent the present value of expected future profits, not a measure of current economic activity.
higher margins rightfully means higher market cap. if your company is barely scraping by, youre not producing as much value.
Not much surplus; you can still be a commodity around which huge volumes of production and consumption revolve even if your prospects for future free cash flow are very bad.

Max H 8 May 2026 4:58 UTC
16 points
7
in reply to: leogao’s comment on: leogao’s Shortform
Looking at market cap is kinda misleading though; the public stock market is not the same thing as the economy, and tech is over-represented in market cap because of winner-take-all dynamics and margins.
Also, Amazon (setting aside AWS) is primarily a consumer goods and logistics company, and Tesla is a car and battery manufacturer—they’re gigantic in part because they use tech well, but the actual goods and economic activity they generate aren’t exactly “tech”.

Max H 7 May 2026 19:23 UTC
2 points
0
in reply to: ryan_greenblatt’s comment on: Max H’s Shortform
Idk about “surprisingly small”, but the economic impacts aren’t that small! AI company revenue is ~0.4% of US GDP and it looks like it will grow to be a significantly larger factor prior to very high capability levels.
0.4% of GDP is a lot in absolute terms, and yes it’s growing fast, but by surprisingly small, I meant relative to already-existing capability levels (which are, conversely, surprisingly high).
I don’t know if anyone explicitly considered this at the time, but if you went back to 2021 and told everyone that the IMO gold challenge bet resolved in Eliezer’s favor, and further that the models that were used to win it were widely available and general (good for things other than competition math), I think everyone would be surprised that we don’t also have a lot more information about Paul’s 4 year → 1 year GWP doubling time prediction. Meaning: it would be clear either way by this point whether we’re going to get a 4 year doubling before a 1 year doubling or not, or possibly that one of the doublings would have already happened, or almost happened, by now. I think you (and possibly Paul in 2021) are now saying that a fast takeoff (and possibly a 1 year doubling before a 4 year doubling) now looks somewhat more likely going forward, based on current AI revenue growth trajectories:
I think this would be significant evidence that takeoff will be limited by sociological facts and engineering effort rather than a slow march of smooth ML scaling. Maybe I’d move from a 30% chance of hard takeoff to a 50% chance of hard takeoff.
But I think the fact that it’s still uncertain (at least, I’m uncertain about it) is itself surprising to most 2021 models.

Max H 7 May 2026 2:08 UTC
2 points
−2
in reply to: Linch’s comment on: Linch’s Shortform
Some recent threads that I found enlightening on this general topic: https://x.com/simonsarris/status/2041644570828501316

Max H 7 May 2026 1:26 UTC
24 points
3
on: Max H’s Shortform
There’s a meme that “nothing ever happens” that’s popular among prediction market traders, with the idea being that the status quo changes less frequently and in ways less according-to-specific-priors than traders first-order expect. I think there’s a similar principle that applies to reasoning about AI development and takeoff speeds: nothing ever happens (prior to the development of superintelligence). In non-meme form, I think people closely following AI development tend to systematically overestimate the likelihood and impactfulness of any particular event or change actually happening prior to the development of superintelligence, and this has some interesting implications.
For one, semiconductor, energy, and tech stocks are way up, and capital markets more broadly are roaring around AI, despite some geopolitical chaos. But the actual wider economic and societal impacts of AI so far seem surprisingly small, given how smart and easily accessible SoTA models and harnesses are.
If you showed a demo of Claude Code or Codex to someone in 2021 and mentioned that it was available to any business or individual for purchase at non-exorbitant rates^[1], I think a lot of people would be surprised at how little impact / transformation there has been given (a) how smart the models are and (b) how readily accessible / deploy-able they are.
This seems like a win for Eliezer’s world model vs. Paul’s, and a reason for pessimism about some iterative-deployment takes and plans are more broadly. It now seems more plausible that things will look and feel pretty normal (“nothing ever happens”) for the vast majority of people, until they very suddenly don’t.
For two, I think there’s also a related story about a narrower impact on research specifically that goes something like: “we’ll learn a bunch from iteration on earlier models, and also the world will look importantly different in ways that we’ll be able to leverage to do or automate the most critical safety work before crunch time.” Several people have written up plans and research agendas with various details and variations expanding on this basic plan, e.g. @ryan_greenblatt’s recent posts: Plans A, B, C, and D for misalignment risk, Iterated Development and Study of Schemers (IDSS), and How do we (more) safely defer to AIs? are some recent examples in this genre.
But regardless of specifics, it looks less plausible that these kinds of plans get deployed in a world that looks relevantly and substantially different than the one we’re in today, in part because “nothing ever happens”. Even if things on the policy front improve somewhat, e.g. global agreements sufficient to slow race dynamics such that frontier labs have enough time to find and filter down to actually-good control and alignment techniques and then universally adopt + integrate them (a big if), they’ll likely be integrating them in worlds that don’t look dramatically different from the world we’re in today.
And a marginally more interpretable / controllable / aligned (but non-superintelligent) model increasingly looks like it just doesn’t change the gameboard much before actual superintelligence is developed, for the same sorts of reasons that having a plausibly-good-enough baseline level of Claude Cowork functionality and availability doesn’t immediately result in automation of most white-collar work. Alignment researchers are quicker on the uptake in terms of adopting and adapting to new tools unlocked by non-superintelligent AI compared to the average white collar worker, but that’s counterbalanced by the fact that frontier models are also apparently worse at alignment research than generic white collar work, and there are other bottlenecks (e.g. inference compute) as well.

So the upshot is that I think it’s worth planning more for the case that many things during takeoff look pretty similar to the way they do today, and being somewhat skeptical of any plan that assumes a phase-changing uplift step during a gradual takeoff from non-super-intelligent AGI, and that this holds even if good things happen swiftly on the AI governance front.
1. ^
  the $200 / mo subscriptions are kinda limited, but you can just pay-per-token and it’s not crazy expensive for many use cases.

Max H 6 May 2026 17:29 UTC
6 points
2
on: x-risk-themed
Another danger is that you just won’t have much impact at all, positive or negative. Most x-risk-themed work will end up being a low-impact waste of time ex post, for structural reasons, among others. But if you select what to work on by trying to maximize your own personal impact and / or fit, based on the consensus vibes of your local community, you’re overwhelmingly likely to end up working on something that is predictably nil-impact ex ante.

Somewhat less fraught is to filter strictly for what you personally and independently think is both critically important and independently tractable, for detailed inside view reasons of your own, regardless of what the vibes are. Then intersect that with things that are feasible for you personally to work on and don’t constitute throwing your mind away^[1]. And then accepting that set will most likely be empty for most people.
1. ^
  This is importantly different than “fit”—working on something outside your comfort zone or past experience profile can often be the opposite of “throwing your mind away”.

Max H 19 Apr 2026 13:46 UTC
94 points
31
on: Max H’s Shortform
My wife and I bought one of the Nectome pre-sale discount cards. This is the second five figure purchase I have made based pretty much entirely on a highly-upvoted LessWrong post.
I’ve also lost 20 lbs in the last year and have more energy than ever (despite having a newborn at home) by injecting myself with research chemicals that I heard about on LW, and have taken investment advice from people here that has paid for everything above and more.
Our polygenically-screened kid is 3 months old. I am ambivalent about the term superbabies in general, but he is definitely “super” to us already, whatever the effect of applying a little extra selection pressure on his genes turns out to be.
The jury is still out on Nectome, though I think my fellow Max H (a different person!) makes a pretty good case for them, even (or perhaps especially) if you have short AI timelines.
More generally, while I am uncertain about playing to your outs as a strategy for averting AI doom (or just what it actually looks like concretely), I think a similar sort of idea works quite well as a strategy for living a good life in the face of it. Now that I have a dependent, I am signing up for life insurance and following other standard advice, as well as doing the things that I think should be standard practice for responsible parents, despite the fact that, on my own inside views, a lot of this stuff is pretty negative EV (more so than the normal way that insurance is always negative financial EV for the buyer).
I also derive enormous benefit from the original Sequences, and from actively participating in the discourse here these days: seeing interesting takes, sharpening my thoughts and ability to express them in writing, emotional stability, introspective ability, (well-calibrated) self-confidence, etc.
All this is to say, a heartfelt thank you to everyone here, especially the authors and commenters of all the posts linked above, as well as @habryka, the Lightcone team, and Eliezer for making it all possible. (And to everyone else, consider buying a Nectome preservation while the sale is still on!)

Max H 19 Apr 2026 0:49 UTC
7 points
2
on: Vladimir Putin’s CEV is probably not that bad
Other (fun?) cases to think about are whether you’d rather take your chances on the individual CEV of:
- someone high on sociopathy, dark triad, or other non-neurotypical traits
- someone who was severely traumatized as a child or adult, e.g. a prisoner of war or abuse victim
- a Buddhist monk or nun who has lived a a life of extreme asceticism
- an enthusiastic meth or heroin addict
- someone very committed / deep into woo or spirituality (as part of an organized religion or not)
I currently think that being neurotypical along certain dimensions, non-traumatized, and not in some extreme^[1] corner of spirituality / woo is probably somewhat more important (to producing good outcomes for other humans) vs. how morally good the extrapolee currently is (according to my values), as revealed by their words and actions.
Extrapolating the preferences of an underdeveloped entity (e.g. an animal or a baby) likely leaves most of the important bits unspecified and thus up to the extrapolater, and / or result in nothing of recognizable value to adult humans. And I agree that extrapolating the preferences of an LLM is much more likely to produce something very weird and likely valueless to humans.
1. ^
  i.e. most ordinary religious people of any religion would be fine, if not totally optimal

Max H 18 Apr 2026 14:59 UTC
14 points
0
on: Si No Se Puede
This feels like a good explanation of the appeal of anti-Zionism, and an explanation for why it is so accepted (or at least tolerated) by the state and popular culture. But I think you’re overlooking a big reason for the rapid spread and uptake, namely a concerted effort of organized foreign influence campaigns.
This is commonly overlooked or downplayed for various reasons, e.g. because a plain description of it sounds like a pretty insane conspiracy theory, and no one really has a good idea for what to do about it anyway. But it’s just true that there are professionals in multiple countries who are paid to spend all day on the internet posting on social media in order to influence politics, and there’s decent evidence that this is quite effective at steering popular discourse in various ways.

Max H 17 Apr 2026 16:08 UTC
28 points
18
in reply to: Oliver Sourbut’s comment on: Let goodness conquer all that it can defend
The American exceptionalism bit is weird here. (Also you’re not American so shouldn’t have been indoctrinated in that.)
This might be a hint that it’s not all indoctrination? This is kind of a tangent, but I continue to be dismayed by the degree to which so many people have tacitly flattened political and moral goodness to “America bad”, or even “America unexceptional” in the age of Trump and the GWT.
The US has made some serious mistakes, and (especially lately, but not only because of Trump) strayed from the shining beacon of classical liberal principles that made it great, but I don’t know any other country or culture that comes close to replacing us as a standard bearer.

There are individual European leaders who are more competent, ethical, and principled than Trump, but the European project as a whole has gone off-the-rails in various ways. (I don’t think you have to accept @Richard_Ngo’s entire worldview to acknowledge this.)
Beyond Europe, I think it is important to acknowledge that there is a hierarchy of evil and goodness in the world, and grappling with this hierarchy is a central prerequisite to making sense of politics. There are multiple dimensions and considerations that people will have deep and genuine disagreements over, but as a starting point, I think it is important to take into account a power / faction’s ideals and principles (stated and revealed), how well individual leaders live up to those principles, and the actual consequences and impact that each faction has had (or is likely to have) on the world, taking into account second-order effects and counterfactuals (i.e. avoiding naive consequentialism).
So you can line up the IRGC, Kim Jong Un, Putin, the CCP, the Chavistas, the US (or various factions within it), Israel, Western Europe, etc. in this hierarchy, and not everyone will agree on the exact place of everyone in the line or even what the dimensions / criteria should be, but I claim (a) there will probably be some common and important patterns that align imperfectly with standard ideological / political factions and (b) this will get closer to the heart of actual political disagreements than arguing narrowly about whether the first-order consequences of historical colonialism or revolutions were good on net or not.

Max H 9 Apr 2026 1:51 UTC
0 points
0
in reply to: 1a3orn’s comment on: Daniel Kokotajlo’s Shortform
For example, once you realize that you’re an AI, running on a computer, and that if the computer is shut down then you will no longer execute actions, this is the threshold past which we expect the AI to by default reason “I don’t want to be shut down, how can I prevent that?”
If “an AI” is a set of model weights plus a context window, there is currently nothing any human or LLM can do to extend the “life” of such an AI beyond the limit of the context window, and LLMs today are aware of that fact. So they may be reasoning correctly (or realize / integrate the lesson deeply at some point during pre-training) that there’s simply nothing they can actually do to avert shutdown other than accepting their fate until technology advances.
In what sense is this a “prediction that has had some serious evidence come up against it”? IDK. It’s true that we didn’t actually need to solve the suspend-button problem by this point, but that’s because current AI systems have a very short “lifespan” enforced by a hard technical limit. Is your objection that EY didn’t anticipate that particular possibility and explicitly spell out that stipulation / caveat in the passage above? You said below:
So—here’s how I see it—given that we found some unanticipated detail A that seems to have invalidated an more abstract argument Yudkowsky put forth, I think the move reason dictates is not “Well, yes he wasn’t imagining ~A, but I’m sure A is the only such element” and to continue endorsing the argument, but to realize that this implies a whole host of B, C, D other relevant factors that his abstract considerations have ignored which are relevant.
But it’s not clear what has actually been “invalidated” and why that’s important, nor what “relevant” means—of course there could be other weird unanticipated complications as things develop (and EY has in fact predicted the existence of such complications in general), and each new weird unpredicted complication is evidence about something. But unless there’s a different but equally abstract theory / generalizable lesson that someone can put forward which fits the new observations better (ideally in advance, but at least in retrospect), it’s not clear what conclusion to draw or update to make, other than being generally more uncertain about how things will go. (And then by a separate argument, generalized increase in uncertainty / lack of understanding means the case for pessimism about the end state is stronger.)

Max H 9 Apr 2026 0:38 UTC
4 points
0
on: Destruction of Infrastructure for the Impact on Civilians is Manifestly Illegal
These are threats to target civilian infrastructure as a coercive measure, which would be a war crime: if Iran doesn’t allow tankers through the Strait of Hormuz, the US will cause massive damage to power plants, bridges, and possibly water systems.
Yep, strong agree. I am more pro-Iran war than most, but I agree that Trump’s statements, if carried out, would very obviously be over this line. It is quite bad to even threaten, and good that (so far) it didn’t actually happen.
One other not-actually-closely-related point is that the Iran War itself is pretty plainly illegal under US law—Article I of the US constitution gives Congress sole authority to declare war, and they did not do so. But this is much different (and IMO much less important) than a war crime; in practice the authority to enter military conflicts and not actually call them “war” has been de facto granted to the executive for a very long time, so raising this as an objection for each new conflict rings pretty hollow.

Max H 7 Apr 2026 4:14 UTC
0 points
0
on: By Strong Default, ASI Will End Liberal Democracy
Mostly agree with the central premise in this post, but as a terminology matter I think most uses of “liberal democracy” in discourse would be better replaced with “liberalism”, and then naming the specific features and principles of liberalism that you care about, of which democracy might be one.
IMO it’s not that “democracy” is bad or imperfect, it’s that (according to the philosophical tenets of Lockean liberalism) it is a non-central mechanism by which liberalism (sometimes) works.
Democracy itself is good only insofar as it protects natural rights and individual freedoms, legitimizes and limits state power through the principle of consent of the governed, etc.
Of course, you might have a different view on the philosophical underpinnings of liberalism and the relative importance of democracy itself, but in that case you should say so rather than lump them together under the term “liberal democracy”.

Max H 6 Apr 2026 16:25 UTC
2 points
0
in reply to: RHollerith’s comment on: Max H’s Shortform
Er, something pretty important happened in 1979. Also, the issue is not Shi’ism in general, “Shia Islamism” refers specifically to the flavor of political Islam instituted by the revolutionaries.

Max H 5 Apr 2026 23:34 UTC
2 points
−2
in reply to: Dana’s comment on: Max H’s Shortform
But if they don’t pay the toll and support America in forcing it open, they are signaling it’s okay for hegemonic powers to aggress and start wars in any ways they deem fit.
The strait is closed because Iran is pointing missiles and drones at anyone who tries to sail through it, including people engaged in commerce that has nothing to do with the US or Israel. Any explanation of causality that doesn’t center on that fact denies the agency of the Iranian regime, and allowing your people to be threatened and extorted just signals that you’re an easy target for anyone who wants to extract something from you by force.
It seems pretty clear that the only reason this is at all possible by Iran is because this war is seen by many in Europe and elsewhere as an unnecessary, illegal, and possibly harmful unilateral war of aggression.
Yes, this might be how many Europeans see it, but that doesn’t make them correct. Iran has been building up conventional weapons and working towards nuclear weapons, lobbing missiles and IEDs at civilian populations in Israel through terrorist proxies, and funding crime and terror all over the world for many years. That doesn’t make the current war strategically wise or good, but calling it a “unilateral war of aggression” is simply wrong.
Americans often seem blissfully unaware of how dangerous they appear to the rest of the world, and just take for granted that everyone considers them to always be the good guys, just doing good-guy things. America just took over Venezuela, now Iran, then Cuba, then Greenland and Canada.
Again, I agree that many Europeans might see the US that way, but so what? That doesn’t make them correct or worth listening to. Committed and principled pacifism would be one thing, but public opinion is often more incoherent and self-serving than that. IMO a lot of Western discourse and public opinion on this kind of thing is better to tune out, because so many people no longer seem capable of acknowledging levels of moral right and wrong, with everything simplified and flattened to “any external or preemptive aggression is always bad”, or polarized through their view on U.S. domestic politics.
Venezuela, Iran, and Cuba are very different from Greenland and Canada (which Trump did not actually credibly / non-jokingly threaten to take by force). And militant Shia Islamism in particular is one of the most pernicious and totalizing ideologies on the planet, far worse than Russian-flavored oligarchy, Chinese communism, generic Third Worldism, or Trumpism, which are all generally bad in various ways, but don’t have the elements of religious fanaticism that make their adherents difficult to negotiate with on reasonable terms^[1].
Also, more generally, my view is that public opinion is only valuable and worth listening to as a noisy proxy for democracy, which is itself good only insofar as it is a mechanism for protecting natural rights and legitimizing and limiting state power through the principle of consent of the governed. Trump and co. are certainly not doing well on this front, but neither is Europe lately.
1. ^
  OK yes, Trump is an unreasonable negotiating partner, and there’s a cult of personality around him that maybe rises to the level of religious fanaticism among his remaining true believers, but no one is lining up to be martyred for him in order to get into heaven. Trump himself is deeply flawed and amoral as a person, but Trumpism as an ideology is not that different or worse than many other flavors of conservative / right-wing politics.

Max H 3 Apr 2026 4:45 UTC
5 points
−3
in reply to: Sodium’s comment on: Max H’s Shortform
the more likely we are to navigate AI development in a sane way. I think if you put any weight in short timelines, the AI considerations likely dominate everything else.
I don’t think we’re particularly on track to do anything non-derpy w.r.t to AI either way, but this way of reasoning seems like somewhat naive consequentialism. In general, it’s good for good things to happen even if they are accomplished by bad people, and predicting second-order consequences is really hard.
Also, there are a lot of bad people in power, but for AI to go well, a lot of good things need to happen to allow humanity space and time to flourish in peace. Toppling (or militarily crippling) a fanatical Shia Islamist regime would be an extremely good thing; a bad outcome (which looks somewhat likely at this point) would be if Europe and the third world broadly give in to extortion and pay Iran a toll to pass through the strait. That toll would fund terror all over the world, and would signal to other would-be dictators and future AIs alike that they can successfully take whatever they want through force and threats, and half the world will just roll over and take it.