Randaly

Karma: 2,108

Randaly Apr 24, 2025, 3:35 AM
2 points
0
in reply to: Vladimir_Nesov’s comment on: Comments on “AI 2027”
You seem to be assuming that there’s not significant overhead or delays from negotiating leases, entering bankruptcy, or dealing with specialized hardware, which is very plausibly false.
If nobody is buying new datacenter GPU’s, that will cut GPU progress to ~zero or negative (because production is halted and implicit knowledge is lost). (It will also probably damage broader semiconductor progress.)
This proportionally reduces cost of inference (and also of training).
This reduces the cost to rent a GPU-hour, but it doesn’t reduce the cost to the owner. (OpenAI, and every frontier lab but Anthropic, will own much or all^[1] of their own compute. So this doesn’t do much to help OpenAI in particular.)
I think you have a misconception about accounting. GPU depreciation is considered on an income statement, it is part of the operating expenses, subtracted from gross profit to get net profit. Depreciation due to obsolescence vs. breakdowns isn’t treated differently. If OpenAI drops its prices below the level needed to pay for that depreciation, they won’t be running a (net) profit. Since they won’t be buying new GPU’s, they will die in a few years, once their existing stock of GPU’s breaks down or becomes obsolete. To phrase it another way, if you reduce GPU-time prices 3-5x, the global AI compute buildout has not in fact paid for itself.
1. ^
  OpenAI has deals with CoreWeave and Azure; they may specify fixed prices; even if not, CoreWeave’s independence doesn’t matter here, as they also need to make enough money to buy new GPU’s/repay debt. (Azure is less predictable.)

Randaly Apr 21, 2025, 2:56 AM
2 points
0
in reply to: Vladimir_Nesov’s comment on: Comments on “AI 2027”
Thanks for explaining. I now agree that the current cost of inference isn’t a very good anchor for future costs in slowdown timelines.
I’m uncertain, but I still think OpenAI is likely to go bankrupt in slowdown timelines. Here are some related thoughts:
1. OpenAI probably won’t pivot to the slowdown in time.
  1. They’d have < 3 years to do before running out of money.
  2. Budgets are set in advance. So they’d have even less time.
  3. All of the improvements you list cost time and money. So they’d need to continue spending on R&D, before that R&D has improved their cost of inference. In practice, they’d need to stop pushing the frontier even earlier, to have more time and money available.
  4. There’s not many generations of frontier models left before Altman would need to halt scaling R&D.
  5. Altman is currently be racing to AGI; and I don’t think it’s possible, on the slowdown hypothetical, for him to get enough evidence to convince him to stop in time.
2. Revenue (prices) may scale down alongside the cost of inference
  1. Under perfect competition, widely-shared improvements in the production of a commodity will result in price decreases rather than profit increases.
  2. There are various ways competition here is imperfect; but I think that imperfect competition benefits Google more than OpenAI. That’s really bad, since OpenAI’s finances are also much worse.
  3. This still makes the cost of inference/cost of revenue estimates wrong; but OpenAI might not be able to make enough money to cover their debt and what you called “essential R&D”. Dunno.
3. Everybody but Anthropic are already locked in to much of their inference (and R&D) spending via capex.
  1. AI 2027 assumes that OpenAI uses different chips specialized in training vs. inference.
  2. The cost of the datacenter and the GPU’s it contains is fixed, and I believe it makes up most of the cost of inference today. OpenAI, via Project Stargate, is switching from renting GPU’s to building its own datacenters. (This may or may not bring down the cost of inference on its own, depending on what kind of discount OpenAI got from Azure.)
  3. So for inference improvements to be a pure win, OpenAI’s usage needs to grow to the same extent. But in this scenario capabilities have stopped improving, so it probably won’t. Usage might even shrink if some people adopt AI in preparation for future capability improvements.
  4. How to account for capex seems complicated:
    We need to amortize the cost of the GPU’s themselves. GPU’s in heavy usage break down over time; OpenAI would need to make enough of a profit to repay the loans and to buy replacement GPU’s. (If they have lower utilization, their existing stock of GPU’s will last longer.) Also, if semiconductor progress continues it will make their existing GPU’s obsolete. I’m skeptical that semiconductor progress will continue at the same rate; but if it does, OpenAI will be locked into using obsolete and inefficient GPU’s.
    Project Stargate is planning on spending 100 billion at first, 50 billion of which would be debt. CoreWeave is paying 11-14% annual interest on its debt; if OpenAI pays 10%, it would be locked into paying ~5 billion/year in interest. This, plus essential R&D, inference electricity, revenue share, and other costs, might or might not be doable; however, they’d almost certainly not have much money left over to buy new GPU’s, save up for a balloon payment, or make a profit.

Randaly Apr 21, 2025, 2:09 AM
2 points
0
on: Finance and AI Timelines
I agree that finances are important to consider. I’ve written my thoughts on them here; I disagree with you in a few places.
(1) Given Altman’s successful ouster of the OpenAI board, his investors currently don’t have much drive/desire/will to force him to stop racing. They don’t have much time to do so on the current pace of increasing spending before OpenAI runs out of money.
(2) It’s not clear what would boost revenue that they’re not already doing; the main way to improve profits would just be to slash R&D spending. Much of R&D spending is spent on research compute; OpenAI is intending to own its own datacenters, so it’s not clear whether they meaningfully can switch paths quickly.
(3) OpenAI is at a massive structural disadvantage to the rest of the frontier companies: they send 20% of their revenue to Microsoft, and they’re taking on tens of billions in debt, which will need to be repaid with interest. So it’s unlikely that they’ll ever be profitable.

Randaly Apr 17, 2025, 8:16 PM
2 points
0
in reply to: niplav’s comment on: niplav’s Shortform
What prompts did you use? Can you share the chat? I see Sonnet 3.7 denying this knowledge when I try.

Randaly Apr 15, 2025, 6:07 AM
5 points
1
in reply to: Vladimir_Nesov’s comment on: Comments on “AI 2027”
I want to clarify that I’m criticizing “AI 2027”’s projection of R&D spending, i.e. this table. If companies cut R&D spending, that falsifies the “AI 2027″ forecast.
In particular, the comment I’m replying to proposed that while the current money would run out in ~2027, companies could raise more to continue expanding R&D spending. Raising money for 2028 R&D would need to occur in 2027; and it would need to occur on the basis of financial statements of at least a quarter before the raise. So in this scenario, they need to slash R&D spending in 2027- something the “AI 2027” authors definitely don’t anticipate.
Furthermore, your claim that “they are losing money only if you include all the R&D” may be false. We lack sufficient breakdown of OpenAI’s budget to be certain. My estimate from the post was that most AI companies have 75% cost of revenue; OpenAI specifically has a 20% revenue sharing agreement with Microsoft; and the remaining 5% needs to cover General and Administrative expenses. Depending on the exact percentage of salary and G&A expenses caused by R&D, it’s plausible that OpenAI eliminating R&D entirely wouldn’t make it profitable today. And in the future OpenAI will also need to pay interest on tens of billions in debt.

Randaly Apr 14, 2025, 9:27 AM
3 points
1
in reply to: elifland’s comment on: Comments on “AI 2027”
My intuitions are more continuous here. If AGI is close in 2027 I think that will mean increased revenue and continued investment
Gotcha, I disagree. Lemme zoom on this part of my reasoning, to explain why I think profitability matters (and growth matters less):
(1) Investors always only terminally value profit; they never terminally value growth. Most of the economy doesn’t focus much on growth compared to profitability, even instrumentally. However, one group of investors, VC’s, do: software companies generally have high fixed costs and low marginal costs, so sufficient growth will almost always make them profitable. But (a) VC’s have never invested anywhere even close to the sums we’re talking about, and (b) even if they had, OpenAI continuing to lose money will eventually make them skeptical.
(For normal companies: if they aren’t profitable, they run out of money and die. Any R&D spending needs to come out of their profits.)
(2) Another way of phrasing point 1: I very much doubt if OpenAI’s investors actually believe in AGI- Satya Nadella explicitly doesn’t, others seem to use it as an empty slogan. What they believe in is getting a return on their money. So I believe that OpenAI making profits would lead to investment, but that OpenAI nearing AGI without profits won’t trigger more investment.

(3) Even if VC’s were to continue investment, the absolute numbers are nearly impossible. OpenAI’s forecasted 2028 R&D budget is 183 billion; that exceeds the total global VC funding for enterprise software in 2024, which was 155 billion. This would be going to purchase a fraction of a company which would be tens of billions in debt, which had burned through 60 billion in equity already, and which had never turned a profit. (OpenAI needing to raise more money also probably means that xAI and Anthropic have run out of money, since they’ve raised less so far.)
In practice OpenAI won’t even be able to raise its current amount of money ever again: (a) it’s now piling on debt and burning through more equity, and is at a higher valuation; (b) recent OpenAI investor Masayoshi Son’s SoftBank is famously bad at evaluating business models (they invested in WeWork) and is uniquely high-spending- but is now essentially out of money to invest.
So my expectation is that OpenAI cannot raise exponentially more money without turning a profit, which it cannot do.

Randaly Apr 12, 2025, 5:30 AM
4 points
0
in reply to: romeo’s comment on: Comments on “AI 2027”
Thanks for the response!
So maybe I should just ask whether you are conditioning on the capabilities progression or not with this disagreement? Do you think $140b in 2027 is implausible even if you condition on the AI 2027 capability progression?
I am conditioning on the capabilities progression.
Based on your later comments, I think you are expecting a much faster/stronger/more direct translation of capabilities into revenue than I am- such that conditioning on faster progress makes more of a difference.
The exact breakdown FutureSearch use seems relatively unimportant compared to the high level argument that the headline (1) $/month and (2) no. of subscribers, very plausibly reaches the $100B ARR range, given the expected quality of agents that they will be able to offer.
Sure, I disagree with that too. I recognize that most of the growth comes from the Agents category rather than the Consumer category, but overstating growth in the only period we can evaluate is evidence that the model or intuition will also overstate growth of other types in other periods.
I don’t think a monopoly is necessary, there’s a significant OpenBrain lead-time in the scenario, and I think it seems plausible that OpenBrain would convert that into a significant market share.
OpenBrain doesn’t actually have a significant lead time by the standards of the “normal” economy. The assumed lead time is “3-9 months”; both from my very limited personal experience (involved very tangentially in 2 such sales attempts) and from checking online, enterprise sales in the 6+ digits range often take longer than that to close anyways.
I’m suspicious that both you and FutureSearch are trying to apply intuitions from free-to-use consumer-focused software companies to massive enterprise SAAS sales. (FutureSearch compares OpenAI with Google, Facebook, and TikTok.) Beyond the length of sales cycles, another difference is that enterprise software is infamously low quality; there are various purported causes, but relevant ones include various principal-agent problems: the people making decisions have trouble evaluating software, won’t necessarily be directly using it themselves, and care more about things aside from technical quality: “Nobody ever got fired for buying IBM”.

Randaly Apr 11, 2025, 9:45 PM
2 points
0
in reply to: elifland’s comment on: Comments on “AI 2027”

I’d be curious to hear more about what made you perceive our scenario as confident. We included caveats signaling uncertainty in a bunch of places, for example in “Why is it valuable?” and several expendables and footnotes. Interestingly, this popular YouTuber made a quip that it seemed like we were adding tons of caveats everywhere,

I was imprecise (ha ha) with my terminology here- I should have only talked about a precise forecast rather than a confident one, I meant solely the attempt to highlight a single story about a single year. My bad. Edited the post.

Randaly Apr 11, 2025, 12:25 AM
2 points
0
on: METR: Measuring AI Ability to Complete Long Tasks
Typo: The description for table 2 states that “In total, 148 of our 169 tasks have human
baselines, but we rely on researcher estimates for 21 tasks in HCAST.”. This is an incorrect sum; the right figure is 149 out of 170 tasks, per the table.

Randaly Apr 10, 2025, 5:30 AM
6 points
0
on: AI 2027: What Superintelligence Looks Like
Footnote 18 of the timelines forecast appears to be missing its explanation.

Randaly Dec 21, 2023, 2:24 PM
20 points
13
in reply to: gwern’s comment on: Effective Aspersions: How the Nonlinear Investigation Went Wrong
Those were in fact some of the cases I had in mind, yes, thank you—I read the news too. And what one learns from reading about them is how those are exceptional cases, newsworthy precisely because they reached any verdict rather than settling, driven by external politics and often third-party funding, and highly unusual until recently post-2016/Trump. It is certainly the case that sometimes villains like Alex Jones get smacked down properly by libel lawsuits; but note how wildly incomparable these cases are to the blog post that Spartz is threatening to sue over. Look, for example, at Jones’s malicious behavior even during the trial, or to take the most recent case, Giuliani, his repeated affirmation of the libelous and blatantly false claims. Look at the kinds of claims being punished in these lawsuits, like claiming that the Sandy Hook shooting was completely fake and the victims’ relatives are fabricating their entire existence. (You’re going to analogize this to Pace saying ‘Nonlinear may not have treated some employees very well’? Really?) Look at how many of the plaintiffs are private citizens, who are in no way public figures. That this rash of victories for the good guys involves lawsuits does not redeem the general abuse of lawsuit
(1) This is a response to you writing “you can count on one hand the sort of libel lawsuit which follows this beautiful fantasy”. Sarcastically stating “I read the news too” doesn’t help you- how obvious these are just makes it worse! You now seem to have entirely abandoned that standard without changing your mind. I can very easily start listing more libel cases that match the new distinctions you’re drawing, to the extent that they are clear enough; is there any point to me doing so? What is the evidence that would convince you that you’re wrong?
(2) One reason I’m confident that you don’t care about the distinctions you’re drawing is that the cases I cited already meet some of the standards you’ve now proposed, and you didn’t care enough to check. In particular, you wrote that “those are exceptional cases, newsworthy precisely because they reached any verdict rather than settling”. This is false, and you provided no justification or evidence for it. I cited six cases; none of them have reached verdicts. Dominion v. Fox, Khalil v. Fox, and Coomer v. Newsmax all did settle; Smartmatic v. Fox and Andrews v. D’Souza are still outstanding and so hasn’t reached a verdict. Weisenbach v. Project Veritas doesn’t appear to have been updated since 2022, but there has been no verdict that I can find. (Given that I’ve already presented you with cases satisfying one of the distinctions you drew above, are you now convinced that you were wrong?)
***
None of which Ben Pace has done, and which is part of why I say he would have excellent odds: it’s unclear what damage has been done to Nonlinear, he had good grounds for his claims, has not said anything which would rise to the level of ‘actual malice’ against a public figure like Spartz, and was “well-intentioned”.
it’s unclear what damage has been done to Nonlinear
Do you believe that unclear damages mean that you can’t win a lawsuit? If so, that’s untrue; damages are often also in dispute. (Did you mean to claim that there was zero damage done? That is different from what you wrote, and is false.)
he had good grounds for his claims
The legal term relevant here is “negligence”; “good grounds” is not the relevant legal terminology. He was negligent in publishing without giving Nonlinear time to reply or updating based on Spencer Greenberg’s evidence; in particular, Habryka stated that they had received evidence that claims in the post were false before they published. [ETA: Habrya comments on this here.] Why do you believe that this wasn’t negligent, if that’s what you meant by writing “had good grounds for his claims”? Or did you mean something else?
has not said anything which would rise to the level of ‘actual malice’ against a public figure like Spartz, and was “well-intentioned”.
You haven’t explained why you think that Spartz is a public figure; again, I find your lack of clear reasoning frustrating to deal with. In this specific case, searching for comments on ‘public figure’ by you in an attempt to figure out what you were thinking, I found a comment by you which did explain your reasoning:
He [Emerson Spartz] very obviously is one [a public figure]. As habryka points out, he has a WP entry backed by quite a few sources about him, specifically. He has an entire 5400-word New Yorker profile about him, which is just one of several you can grab from the WP entry (eg. Bloomberg). For comparison, I don’t think even Eliezer has gotten an entire New Yorker profile yet! If this is not a ‘public figure’, please do explain what you think it would take. Does he need a New York Times profile as well? (I regret to report that he only has 1 or 2 paragraphs thus far.)
Now, I am no particular fan of decreeing people ‘public figures’ who have not particularly sought out fame (and would not appreciate becoming a ‘public figure’ myself); however, most people would say that by the time you have been giving speeches to universities or agreeing to let a New Yorker journalist trail you around for a few months for a profile to boost your fame even further, it is safe to say that you have probably long since crossed whatever nebulous line divides ‘private’ from ‘public figure’.
Even in that comment, you never actually stated what you believe the standard for being a public figure is, or gave any legal citations to support that standard.^[1] However, there’s at least enough detail to say that your claim is wrong; it’s absolutely not true that having a magazine profile is the level of fame required to make you a public figure. Waldbaum v. Fairchild Publications describes the standard for general public figures^[2] as follows:
a person can be a general public figure only if he is a “celebrity”—his name a “household word”—whose ideas and actions the public in fact follows with great interest
To give a concrete example, musician Dr. Luke owns two publishing companies, has been nominated for numerous Grammys, and has had plenty of magazine articles written about him, including one specifically in the New Yorker; courts have also repeatedly held that he isn’t a general public figure.
***
This is a double bait and switch: Nonlinear is just Spartz who can spend his money on whatever he likes, while the ‘EA community’ is not the one being specifically threatened, and even comparing Nonlinear and Lightcone is misleading, because Lightcone has many other ongoing responsibilities (such as spending money to maintain the website this is being written on or renovate buildings) and it would presumably be Pace personally being sued as well.
My comment was not misleading. I was explicitly responding to a quote, which I directly quoted in my comment right above what you responded to, where you stated that “[lawsuits] are cynically used to burn money based on the fact that rich people have a lot more money than poor people”. This is about “rich” vs. “poor”. The rest of the quote is “money has log utility, so when they burn $10k+ to burn your $10k+, they come out way ahead in punishing you, and the goal is not even to win the lawsuit, it’s to force you to a settlement when you run out of money. Which you will before they do, and in the case of Spartz suing Lightcone, it’s not like Lightcone has a ton of idle cash they can burn on a defense of their claims”.^[3]
Some libel threats fall under your analysis; others do not. I have already given many examples of lawsuits that do not- Fox News, for example, is not likely to simply run out of money, nor is it poorer than its various legal opponents. Your analysis of this specific case is wrong; Habryka has explicitly stated that Lightcone has enough money to defend a libel suit. He also said that Lightcone would probably be able to fundraise from the EA community for a defense.
***
I found this extremely frustrating to reply to. I personally regard most of the concrete claims you made in the original comment as being not just wrong, but both obviously wrong and unsupported. You seem to have abandoned actually defending them, and indeed even noted how obvious my counterarguments were- “yes, thank you—I read the news too”. (You didn’t change your mind in response, though!)
Judging by the timestamps, you wrote your long response very quickly. It’s taken me much, much longer to write this reply, and I’m only a small fraction through replying so far. (I’ll reply to the rest later, I guess, ugh ugh ugh.) There’s a very obvious reason why you were so much faster: you didn’t bother to defend your specific previous claims or to check if the new stuff you tossed out was actually right. It would have taken you ~10 seconds to verify if any of the lawsuits named reached a verdict, instead of wrongly making up that they all had; it’s taken me much longer to check all of them and write up a reply. It would have taken you ~5 minutes^[4] to find the legal definition of public figure, instead of making up your own. It’s taken me far longer to find a different comment that actually explained what you were talking about, and to then lookup and write a response myself, including even finding a specific person who was both the subject of a New Yorker article and had been determined to not be a public figure. This is a gish gallop.
1. ^
  The linked comment is in response to a comment by an attorney who correctly stated the standard for being a public figure and correctly stated that Spartz isn’t a public figure… which you ignored when you made up your own uncited standard for what a public figure is. (It’s also a pretty devastating indictment of LW that the attorney commenting with a correct definition and application of “public figure” received considerably less karma/agreement than you making up your own incorrect standard, which gave a more popular answer.)
2. ^
  There’s also the category of “limited purpose public figure”. Spartz also probably (but not definitely) isn’t one; all of the citations you gave- and probably almost all of his publicity, judging by his Wikipedia page- don’t relate to Nonlinear or AI broadly, or their treatment of interns specifically.
3. ^
  The new argument that you’ve made here might or might not be true; you’ve tossed it out without sufficient justification. Nonlinear would also like to spend money on other things, and I don’t know how to compare their resources, preferences, and alternative expenditures vs. Lightcone; you haven’t even tried. (Note that your argument requires a significant difference.)
4. ^
  Arguably, it would have taken you zero minutes; the comment I linked was a response to an attorney who told you the correct answer.

Randaly Dec 19, 2023, 11:41 PM
54 points
34
in reply to: gwern’s comment on: Effective Aspersions: How the Nonlinear Investigation Went Wrong
This is a combative comment which fails to back up its claims.

how surely only noble and good people ever sue over libel

if you really believe lawsuits are so awesome and wonderful

He did not say this. This is not reasonable for you to write.

you can count on one hand the sort of libel lawsuit which follows this beautiful fantasy.

This is not true. This is obviously not true. A successful and important libel case (against Giuliani) was literally headline news this week. You can exceed five such cases just looking at similar cases: Dominion v Fox; Smartmatic v Fox; Coomer v Newsmax; Khalil v Fox; Andrews v D’Souza; and Weisenbach v Project Veritas. This is extremely unreasonable for you to say.

They are cynically used to burn money based on the fact that rich people have a lot more money than poor people

Nonlinear certainly doesn’t have more money than the EA community. Nonlinear plausibly (?) doesn’t have more money than Lighthouse; at a minimum, it’s not a significant difference.

which they would generally win, BTW

<argument needed>

It’s very unclear to me whether Lighthouse would win; your confidence here seems unreasonable; but more importantly, “no, that’s not true” is just not a useful thing to say here. (You’re responding to a post that did have many good citations of cases; seems like most people think it’s plausible they’d lose.)

And a lawsuit is a way to destroy someone, not counter-argue them.

In the most blandly literal sense possible, lawsuits are arguments.

what goes on inside a court has only a questionable relationship to counterargument to begin with, which is why a decent chunk of rationality is about explaining why legal norms are so inappropriate for rational thinking

You have again not given any argument for this.

The rules under which lawsuits proceed are deliberately setup in an attempt to get at the truth. Specific requirements- from the prohibition on hearsay; to the requirement of a neutral and unbiased jury; to the requirement that both sides be able to examine and respond to evidence and arguments- are both truthseeking and not generally followed outside of the court system.

“My ingroup’s internet discussions are so great that they’re not only better than the outside society’s way of determining contested questions, they invalidate their use” is a dangerously culty belief. I think it is particularly bad in this context, since the initial post had specific failures that the legal system would have handled correctly. (eg not giving Nonlinear time to respond; it’s possible that I’ll feel like the eventual outcome here is reasonable, IDK, but the initial post had clear issues.) But at a minimum, if you’re saying that people be “shunned, demonized, and criticised” (!), you really ought to say specifically why/how the courts would be unreliable in this case.

Randaly Sep 8, 2023, 5:24 AM
9 points
−3
in reply to: habryka’s comment on: Sharing Information About Nonlinear
IMO, both U.S. and UK libel suits should both be very strongly discouraged, since I know of dozens of cases where organizations and individuals have successfully used them to prevent highly important information from being propagated, and I think approximately no case where they did something good (instead organizations that frequently have to deal with libel suits mostly just leverage loopholes in libel law that give them approximate immunity, even when making very strong and false accusations, usually with the clarity of the arguments and the transparency of the evidence taking a large hit).
(Unavoidably political, as lawsuits often are)
A central example of the court system broadly, and libel lawsuits narrowly, promoting better epistemics are the allegations that the 2020 election was fraudulent.
It is certainly not true that there are always loopholes that give immunity, see e.g. Fox News’ very expensive settlement in Dominion v. Fox News.
More broadly: “Trump, his attorneys, and his supporters falsely asserted widespread election fraud in public statements, but few such assertions were made in court.” The false allegations of fraud were dependent on things like hearsay, false claims that opponents weren’t given a chance to respond to, and vague or unsupported claims; virtually all discussion on the internet, and this post in particular, feature all three; the court system explicitly bans these. (Note that people who can’t support their case under legal standards of evidence often just settle or don’t bring a case in the first place.)

Randaly Sep 1, 2023, 10:18 PM
7 points
1
in reply to: Bird Concept’s comment on: A Golden Age of Building? Excerpts and lessons from Empire State, Pentagon, Skunk Works and SpaceX
I’m not disputing that specific people at Skunk Works believed that their tech was disliked for being good; but that’s a totally insane belief that you should reject immediately, it’s obviously self-serving, none of those people present any evidence for it, and the DoD did try to acquire similar technology in all these cases.
Again, this is a direct quote on procurement incentives from a guy who was involved on both the buy and sell side of the SR-71 back in the day.
This is quote from, per you, somebody from the CIA. The CIA and Air Force are different organizations; he was presumably not involved in the Air Force’s decision not to acquire the F-12B. We have definitive proof that the Air Force’s procurement decisions weren’t necessarily opposed to high performing planes, since they had planned on acquiring different, but similarly capable, planes.
There’s also a more gnarly philosophical issue here, in terms of the “insane” belief you’re pointing to. I find it fairly plausible that individual commanders might have incentives that are different from those of the Navy, or the Air Force, as a whole, and that this might drive procurement decisions.
I am very confident that the book-length sequence you linked to doesn’t contain a justification for the claim that “individual Air Force commanders hate fast planes”. But if it does, please provide the actual justification instead of linking to a ~150 page book (“go read the sequences”).

ETA: I may have misunderstood your point; if you instead meant literally just to justify the sentence you wrote, that principal-agent problems are possible, then I don’t disagree; that does absolutely nothing to justify the specific claimed principal-agent problem.

Randaly Sep 1, 2023, 12:35 PM
63 points
11
on: A Golden Age of Building? Excerpts and lessons from Empire State, Pentagon, Skunk Works and SpaceX
Your discussion of Skunk Works is significantly wrong throughout. (I am not familiar with the other examples.)
For example, in 1943 the Skunk Works both designed and built America’s first fighter jet, the P80 Shooting Star, in just 5 months. Chief engineer Kelly Johnson worked with a scrappy team of, at its peak, 23 designers and 105 fabricators. Nonetheless, the resulting plane ended up being operationally used by the air force for 40 years.
The P80 was introduced in 1945; the US almost immediately decided to replace it with the F-86, introduced in 1949. The phrase “operationally used by the air force for 40 years” is only technically true because rather than scrap existing P80 production, they were modified slightly and used as training aircraft.
Our ship had a four-man crew — commander, helmsman, navigator and engineer. By contrast, a frigate doing a similar job had more than three hundred crewmen. … Our stealth ship might be able to blast out of the sky a sizable soviet attack force, but in terms of an officer’s future status and promotion prospects, it was about as glamorous as commanding a tugboat. At the highest levels, the Navy brass was equally unenthusiastic about the small number of stealth ships they would need to defend carrier task forces. Too few to do anyone’s career much good in terms of power or prestige.
This is wrong. Their stealth ship wasn’t able to “blast out of the sky a sizable soviet attack force”, or to do literally anything; it was just a testbed for exploring automation and stealth hulls, totally incapable of doing anything. Skunk Works didn’t actually successfully build anything here! (The stealth design was later used on the Zumwalt class of destroyers, which had unrelated issues.)
Not sure where he got the 300 crew figure from? Even beyond the fact that the Sea Shadow wasn’t actually designed to do anything (and so would need more specialized crew to do so), the Sea Shadow was only a tenth of a the size of the frigates it’s being compared with. (The Navy has since tried to use similar automation to reduce the crew of newer ships; the Gerald Ford class of aircraft carriers represent the realistically achievable reduction in crew via automation: 3,200 → 2,600 or so, so ~20%.) (Note that this also trivially falsifies the claim that the Navy rejects automation to reduce crew sizes?)
“The Navy rejected our ship design because it was totally too good, you just gotta believe me, even though we’ve never ever successfully produced ships” is an insane thing for you to accept with zero evidence.
Yet Lockheed could barely sell [the SR-71]. As described by a CIA engineer inside the Skunk Works:
> … But I never gave him much chance to sell a lot of those airplanes because they were so far ahead of anything else flying that few commanders would feel comfortable leading a Blackbird wing or squadron. I mean this was a twenty.-first-century performer delivered in the early 1960s. No one in the Pentagon would know what to do with it.
This is totally wrong. You are again putting forth the insane claim that people rejected Skunk Work’s technology because of how good it was, with zero actual evidence of why the SR-71 wasn’t mass produced.
The SR-71 (Mach 3.3, 85,000 feet) wasn’t significantly better than planned contemporary planes like the B-70 (Mach 3.1, 77,350 feet) or the F-108 (Mach 3, 80,100 feet). Both of those planes were cancelled, because the development of missiles meant that flying higher and faster was no longer a viable strategy; since then, military planes like the the F-18 (Mach 1.8, 50,000 feet), and the F-35 (Mach 1.6, 50,000 feet) have often been lower and slower. This is a deliberate choice: unmanned, one-way missiles can always go faster than a manned plane. Most of the SR-71′s advantages come not from it being inherently better than any possible missile, but from it being faster and higher than the planes early SAM’s were intended to target; mass production and usage in other roles would inherently make this go away.
(To be clear, Skunk Works was successful and build many things successfully; it’s specifically your claims and examples that are wrong. In particular, you left out most of their successful planes like the F-117.)
What links here?
- L Rudolf L's comment on A Golden Age of Building? Excerpts and lessons from Empire State, Pentagon, Skunk Works and SpaceX by Bird Concept (Dec 10, 2024, 4:19 PM; 4 points)

Randaly Nov 28, 2022, 6:26 AM
4 points
1
in reply to: M. Y. Zuo’s comment on: Analysis: US restricts GPU sales to China
No.
I’m pretty negative on how you fail to discuss any specific claim or link to any specific evidence, but you spend your longest paragraph speculating about the supposed bias of unnamed people.
You haven’t really written enough to be clear, but I suspect that you have confused concentration camps with death or extermination camps? Regardless, the recent UN report did pretty specifically support claims of concentration camps- see points 37-57

Randaly Jun 8, 2022, 7:20 AM
2 points
on: Visible Homelessness in SF: A Quick Breakdown of Causes
I also found that, controlling for rents, the partisanship of a state did not predict homelessness (using the Partisan Voting Index)
This is not a useful way of looking at this; homelessness would be almost entirely controlled by city, not state, policies. State partisanship in large part measures not how blue or red the states’ cities are, but rather how urban or rural the state as a whole is.

Randaly Jul 22, 2021, 11:22 PM
2 points
in reply to: MichaelLowe’s comment on: ($1000 bounty) How effective are marginal vaccine doses against the covid delta variant?
This, and the Bahrain/UAE cases, seem more likely to be driven by concerns about whether/how well the Chinese vaccines work?

Randaly Jun 21, 2021, 4:22 AM
4 points
in reply to: johnswentworth’s comment on: Notes on War: Grand Strategy
On the other hand, look at the US wars in Vietnam, Iraq or Afghanistan. The outcomes of these wars were determined much more by political forces (in both of the relevant countries) than by overwhelming force.
Insurgencies aren’t a good comparison for conventional wars like the Nagorno-Karabakh war.

Randaly Jun 21, 2021, 4:14 AM
6 points
on: Notes on War: Grand Strategy
The overall thrust here seems like an application of Clausewitz’s maxim that “war is the extension of politics by other means”. However, the specific politics suggested seem very unrealistic.
- You suggest ways to impact Azerbaijan’s internal politics by targeting harm to specific groups. I see no reason to believe that Armenia had any substantial ability to deal much harm to Azerbaijan at all, so this isn’t relevant. In general, it would be much harder for Armenia to advance to deal significant damage to Azerbaijan’s homeland than it would be to defend.
- Assassinations are practically universally a bad way to change a country’s politics; they usually result in a direct backfire.
- Your advice seems to lack object-level knowledge of the conflict itself; in particular, Azerbaijan is not a liberal democracy; its current leader has been in power for decades, and won ~86% of the vote in the most recent election.
- The second Nagorno-Karabakh conflict was indeed a populist war: the two main causes were racial conflict between Azerbaijan and Armenia, and the result of the First Nagorno-Karabakh war, in which Armenia overran and occupied Nagorno-Karabakh. “We will end racism against us” is not in fact a realistic short-term plan; it would take decades to have relevance. Resolving the conflict over Nagorno-Karabakh would presumably involve returning Nagorno-Karabakh to Azerbaijan; this is not a good plan for retaining Nagorno-Karabakh! More generally, you can usually prevent a war by just giving in, and maybe Armenia should have given that they lost the war. It’s also not politically realistic, given Armenia’s domestic politics.
- Your plans for improving Armenia’s popularity in Azerbaijan (a) wouldn’t stop the war, (b) likely wouldn’t help end the war, and (c) are irrelevant to actually existing war, since they assume that Armenia is overrunning Azerbaijan, occupying their citizens, taking significant numbers of prisoners, etc.
- It is not clear to me how Armenia would go about creating a recession or a famine in Azerbaijan?
- Pretty much all of these plans are underspecified outcomes, not realistic plans. For example, for “we should do propaganda” you haven’t specified what Armenia should have actually done for it to matter. *In practice*, actually Azerbaijan had an *overwhelming* advantage in propaganda, making use of new channels like TikTok and Youtube to quickly disseminate videos of military successes.
So how should Armenia have retained Nagorno-Karabakh? Given that Azerbaijan is about 3 times its size, and that it has substantial oil reserves that can be used to fund military spending, Armenia would have little chance on its own. Even worse, Azerbaijan is supported by their co-ethnics in Turkey, which is vast and wealthy in comparison to both states; Armenia would not realistically have been able to disrupt this relationship.
Armenia would need a powerful patron to counter this. Three options:
- Iran supported Armenia in the First Nagorno-Karabakh War, but developed closer ties with Azerbaijan more recently. I don’t know how realistic blocking this would be, I’m not very familiar with regional politics.
- Russia has also historically backed Armenia, but didn’t intervene until late in the conflict (Russian peacekeepers are now in the region). This was likely for two reasons: first, Russia is already somewhat overstretched, with concerns in ongoing conflicts in Ukraine, Syria, and Libya. Second, Armenia’s 2018 Revolution brought to power a more democratic government that leaned away from Russia and towards the West. The answer here would be for Armenia to become a more dependent Russian client state, although this may have been impossible for domestic political reasons. Armenia will likely now pursue this strategy.
- Armenia has an extensive diaspora in the West, which it used to mobilize political support; using the diaspora and the Armenian realignment towards the West to secure military aid and security guarantees could have been useful, although in practice I don’t think Armenia could have secured anything of much significance.