90% confident it's a reference to the Principia Discordia's SNAFU Principle, that communication is only possible between equals.
An increasing number of such typos has been impacting my satisfaction with DWATV lately. The content is pushed to prod too fast, without enough time for SPaG verification. ("Shouldn't AI be able to fix this?")
It seems like the #1 resource in Frontier AI development, is talented researchers, which XAI does not appear to have at any significant scale nor is it able to get. (The fact that Anthropic which has the highest employee retention rates among labs, the least capital, and the best model really shows this), makes me less concerned about XAI's safety.
It is however, extremely unfortunate that the richest man in the world seems to understand the problem acutely, probably has one of the most high-leverage positions in the world that could influence it, and is instead doing whatever it is that he is doing.
More evidence of Musk’s ignorance. Data Center in space is very stupid. The #1 requirement for data center is cooling. In space there is only radiative cooling. Not as simple as running water cooling pipes through your data center.
My gut reaction is to agree, but I haven't run the numbers. I'm also _very_ skeptical of putting chip manufacturing on the Moon in any time frame that makes a difference to AI. No water (except maybe small amounts in shadowed craters at the poles), no HF source, no organics for the resists... It sounds like the sort of thing one would do _after_ developing full Drexler/Merkle nanotechnology aka atomically precise manufacturing.
Each business line got evaluated separately. But stack them: launch (SpaceX), communications (Starlink), physical labor (Optimus, orchestrated by Grok), intelligence (xAI), chip fab (TeraFab), energy (100GW solar at both Tesla and SpaceX), and the orbital infrastructure where he says most AI will live. One person, every bottleneck.
I think Elon Musk is building toward unilateral control of post-human civilization. I don't think it's accidental.
Then he told us the capstone. He cited The Moon Is a Harsh Mistress, described the plot where a lunar colony bombards Earth until the government submits, and said "I just want to see that thing in operation." A lunar mass driver breaks MAD. Kinetic impactors from lunar distance arrive at 11+ km/s with no intercept capability, and the only launch system that can reach the Moon to respond belongs to the same guy who owns the mass driver.
He cited the book. He described the scenario. He named the outcome. I think we should treat this as real.
His alignment thinking is confused, sure. But his plan doesn't require alignment (he thinks). It requires control of physical bottlenecks.
I'm calling it now: Musk told us his God Emperor plan on camera. We might should take him seriously.
Fun to think about, but the Departments of Defense of the world are simply not going to allow an unrestricted mass driver on the Moon that doesn’t have some kind of Earth-based control system akin to how nuclear missile launches are controlled. They’d see the potential for it right away and build in safeguards.
For those of you who think Zvi is being harsh here, I just want to reinforce that Musk was deeply unimpressive in this interview. I was hoping this style of long form interview would revive the 'good' Musk, away from Twitter and sycophants, but it did not.
I don't just mean that in an affective way; Musk may be a jerk but a lot of the work he's done is very important. I mean Musk genuinely just seemed not that sharp/switched on, in a way that wasn't true historically.
Dwarkesh was an extremely generous interviewer here but Musk didn't respond to tough questions half as well as, for example, Satya Nadella did.
Musk may, genuinely, be a bit unhinged. He seems to take the simulation hypothesis extremely seriously, have complete non sequitor ideas about what the simulation runners might want or react to, is *maybe not joking* about thinking of himself as the player character in this simulation, and pretty clearly has authoritarian fantasies. I didn't think he meant any of this seriously before I listened to this interview.
I whole-heartedly agree. I kept thinking, what a waste of Collison's time to be in this interview. I wish the interview was just between him and Dwarkesh. Then we could hear how an interesting person thinks. Instead, we get Musk living up to the trope that he's the world's smartest 14 yr old boy, but even that elides some of his major deficiencies. It's not just the juvenile boring humor/interests, he just clearly hasn't (can't) think a great many things through.
Elon thinks he's the Player Character, but he's actually one of those dime-a-dozen minor villains who says "We're all doomed unless I do this horribly unethical thing that will certainly save us!" Hopefully the actual PC will show up soon and choose the option to nonviolently show him the error of his ways and redeem him.
(People who think it's worth Zvi's time to manually proofread for typos in these posts are deeply confused. It's not worth his time for *us*, let alone for him! If there's an error that makes his actual meaning genuinely unclear, sure, comment on it, but in general, being unable (or unwilling) to parse typos is, ¿que se dice?, a Skill Issue :).)
Zvi already committed to less emphasis on Speed Premium Uber Alles going forward for 2026, and also I'm confused by the assertion that the typos in this post and elsewhere require "manual" proofreading. It's not MS Word 95 starring Clippy anymore, the automated checks are decently robust now. Part of why I find it puzzling is that Zvi already does such automatic cleaning-up for most pull quotes, Twitter excerpts, etc., such as when other (non-Roon) writers are seemingly allergic to capitalization. Sure, the Pareto optimal number of typos is not zero; one can still note when The Price Is Write Typ-O-Meter(tm) is being turned a bit beyond the median level for a given blog. I obviously value Zvi's time enough to pay for it either way, but must also value my own time and cognitive load as a DWATV completionist reader too. Levels of Friction, Beware Trivial Inconveniences, signal and noise, etc. Measure a stitch in time twice to cut nine pounds of cure once.
(Which is actually a stone tablet brought down from the mountain labeled "Cost-Benefit Analysis")
PS: I'll add that I am bewildered by your assumption that Zvi isn't already using some kind of editing assistant; of course he is, the typos we see are just the ones that don't get caught, for whatever reasons. But even if that weren't the case, setting up a (new) automated system isn't free, so I'd still just be pointing to the same rock.
Is there no way to report this spambot to substack?
I wish
Can you please don't say Grokopedia anymore? It's Grokipedia with an i like Wikipedia. (You even got some links to it wrong when it launched.)
I'm not a fan of Grokipedia, but I hate misspellings of proper names.
"In important senses communication is only possible and he thus has all the authoritarian communication problems"
Is this sentence missing a phrase or two, or am I misunderstanding?
90% confident it's a reference to the Principia Discordia's SNAFU Principle, that communication is only possible between equals.
An increasing number of such typos has been impacting my satisfaction with DWATV lately. The content is pushed to prod too fast, without enough time for SPaG verification. ("Shouldn't AI be able to fix this?")
fair critique but man...let's acknowledge the pace zvi is pushing pieces to keep up. life is coming at us fast
"let's acknowledge the pace zvi is pushing pieces to keep up. life is coming at us fast"
Seconded! Zvi has been handing a superhuman amount of work.
While I have some preferences that differ from his, I DEEPLY respect the work he is doing.
It seems like the #1 resource in Frontier AI development, is talented researchers, which XAI does not appear to have at any significant scale nor is it able to get. (The fact that Anthropic which has the highest employee retention rates among labs, the least capital, and the best model really shows this), makes me less concerned about XAI's safety.
It is however, extremely unfortunate that the richest man in the world seems to understand the problem acutely, probably has one of the most high-leverage positions in the world that could influence it, and is instead doing whatever it is that he is doing.
Podcast episode for this post:
https://open.substack.com/pub/dwatvpodcast/p/on-dwarkesh-patels-2026-podcast-with-edf
Musks end game is trans humanism…everything is in service of transcending human biology.
Humans cannot live on Mars let alone explore the solar systems.
Consciousness transfer is an engineering problem…bio bias makes us blind.
Biology plus electrons…silicon plus electrons…looks like a duck to me.
Add sense perception and persistent memory and it really acts like a duck.
Think about!
More evidence of Musk’s ignorance. Data Center in space is very stupid. The #1 requirement for data center is cooling. In space there is only radiative cooling. Not as simple as running water cooling pipes through your data center.
My gut reaction is to agree, but I haven't run the numbers. I'm also _very_ skeptical of putting chip manufacturing on the Moon in any time frame that makes a difference to AI. No water (except maybe small amounts in shadowed craters at the poles), no HF source, no organics for the resists... It sounds like the sort of thing one would do _after_ developing full Drexler/Merkle nanotechnology aka atomically precise manufacturing.
Nikola Tesla didn't have kids.
Neither did Isaac Newton, George Washington, Adam Smith...
Not to mention all the religious figures his Republican friends revere. Jesus....
He could show some gratitude for the future they helped create for him.
Musk’s comment on mass driver on the moon is confessional. whoever controls that will rule the world. As good as ICBMs.
Each business line got evaluated separately. But stack them: launch (SpaceX), communications (Starlink), physical labor (Optimus, orchestrated by Grok), intelligence (xAI), chip fab (TeraFab), energy (100GW solar at both Tesla and SpaceX), and the orbital infrastructure where he says most AI will live. One person, every bottleneck.
I think Elon Musk is building toward unilateral control of post-human civilization. I don't think it's accidental.
Then he told us the capstone. He cited The Moon Is a Harsh Mistress, described the plot where a lunar colony bombards Earth until the government submits, and said "I just want to see that thing in operation." A lunar mass driver breaks MAD. Kinetic impactors from lunar distance arrive at 11+ km/s with no intercept capability, and the only launch system that can reach the Moon to respond belongs to the same guy who owns the mass driver.
He cited the book. He described the scenario. He named the outcome. I think we should treat this as real.
His alignment thinking is confused, sure. But his plan doesn't require alignment (he thinks). It requires control of physical bottlenecks.
I'm calling it now: Musk told us his God Emperor plan on camera. We might should take him seriously.
Note also where having 14 *acknowledged* offspring (and reportedly a lot more unacknowledged!) might fit into this whole "God Emperor" thing.
Fun to think about, but the Departments of Defense of the world are simply not going to allow an unrestricted mass driver on the Moon that doesn’t have some kind of Earth-based control system akin to how nuclear missile launches are controlled. They’d see the potential for it right away and build in safeguards.
Zvi is increasingly my interface with the world of tech Twitter, now that my immune system can no longer withstand the hazards of the platform itself.
For those of you who think Zvi is being harsh here, I just want to reinforce that Musk was deeply unimpressive in this interview. I was hoping this style of long form interview would revive the 'good' Musk, away from Twitter and sycophants, but it did not.
I don't just mean that in an affective way; Musk may be a jerk but a lot of the work he's done is very important. I mean Musk genuinely just seemed not that sharp/switched on, in a way that wasn't true historically.
Dwarkesh was an extremely generous interviewer here but Musk didn't respond to tough questions half as well as, for example, Satya Nadella did.
Musk may, genuinely, be a bit unhinged. He seems to take the simulation hypothesis extremely seriously, have complete non sequitor ideas about what the simulation runners might want or react to, is *maybe not joking* about thinking of himself as the player character in this simulation, and pretty clearly has authoritarian fantasies. I didn't think he meant any of this seriously before I listened to this interview.
I whole-heartedly agree. I kept thinking, what a waste of Collison's time to be in this interview. I wish the interview was just between him and Dwarkesh. Then we could hear how an interesting person thinks. Instead, we get Musk living up to the trope that he's the world's smartest 14 yr old boy, but even that elides some of his major deficiencies. It's not just the juvenile boring humor/interests, he just clearly hasn't (can't) think a great many things through.
> let him justify his actions and ago
Should this be "and ego"?
Elon thinks he's the Player Character, but he's actually one of those dime-a-dozen minor villains who says "We're all doomed unless I do this horribly unethical thing that will certainly save us!" Hopefully the actual PC will show up soon and choose the option to nonviolently show him the error of his ways and redeem him.
(People who think it's worth Zvi's time to manually proofread for typos in these posts are deeply confused. It's not worth his time for *us*, let alone for him! If there's an error that makes his actual meaning genuinely unclear, sure, comment on it, but in general, being unable (or unwilling) to parse typos is, ¿que se dice?, a Skill Issue :).)
Zvi already committed to less emphasis on Speed Premium Uber Alles going forward for 2026, and also I'm confused by the assertion that the typos in this post and elsewhere require "manual" proofreading. It's not MS Word 95 starring Clippy anymore, the automated checks are decently robust now. Part of why I find it puzzling is that Zvi already does such automatic cleaning-up for most pull quotes, Twitter excerpts, etc., such as when other (non-Roon) writers are seemingly allergic to capitalization. Sure, the Pareto optimal number of typos is not zero; one can still note when The Price Is Write Typ-O-Meter(tm) is being turned a bit beyond the median level for a given blog. I obviously value Zvi's time enough to pay for it either way, but must also value my own time and cognitive load as a DWATV completionist reader too. Levels of Friction, Beware Trivial Inconveniences, signal and noise, etc. Measure a stitch in time twice to cut nine pounds of cure once.
/me points to the rock labeled "Opportunity Cost"
(Which is actually a stone tablet brought down from the mountain labeled "Cost-Benefit Analysis")
PS: I'll add that I am bewildered by your assumption that Zvi isn't already using some kind of editing assistant; of course he is, the typos we see are just the ones that don't get caught, for whatever reasons. But even if that weren't the case, setting up a (new) automated system isn't free, so I'd still just be pointing to the same rock.
This was the first time I listened to a podcast and could hear Zvi's voice in the background the entire time