Sign in

sesquipedalianThaumaturge

@sesquipedalian.bsky.social
143 followers 79 following 446 posts

LG human effective altruist 4/astronautical engineer 1/speculative fiction enjoyer 3 they/them

PostsRepliesMedia
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 13/10/2026
Sure, it would take a couple more torpedos to finish them off once they were crippled. But the 1939 RN is still totally capable of sinking a bunch of Suezmaxes; the Africans would rely on their air arm to destroy the European fleets before they got close, not the resilience of their transports.
010
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 13/10/2026
and an Exocet will impact at a much lower velocity. My guess is it would totally fail to penetrate the main belt of a battleship and probably struggle against other parts of the armor too (based on heavier bombs in WWII being stopped by armored decks sometimes).
000
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 13/10/2026
My understanding is that early ASMs like the Termit did have warheads which were meant to penetrate thick armor, but that caused ships to stop being armored, which lead to modern ASMs having warheads which weren’t designed for that. 165 kg is quite small compared to heavy naval shells,
100
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 13/10/2026
I think it’s quite plausible that you need big salvos and the Africans run out of ASMs before the Europeans run out of capital ships, and they mostly have to rely on aerial bombing.
010
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 13/10/2026
Also, I’m honestly not sure how effective Exocets would be against superdreadnoughts. They’re not designed to penetrate armor, and I don’t know if they can be programmed to target specific parts of a ship in order to avoid the armor belt.
210
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 13/10/2026
I totally agree that the naval theater would be pretty one-sided in Africa’s favor, I’m just objecting to the specific claim about a modern cargo ship being very hard to sink. A Suezmax may be large, but it also doesn’t have armor or anti-torpedo bulkheads.
210
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 13/10/2026
Why would modern cargo ships be very hard to sink with 1939 technology? I don’t think it would take all that many torpedo hits or 15-inch shells to sink one.
110
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 29/09/2026
Whether rationalists are empiricist or not is debatable, but “they specifically contrast themselves with empiricism” is incorrect in my experience. In fact they love talking about how good empiricism is. For example: www.lesswrong.com/posts/7ZqGiP...
lesswrong.com
Twelve Virtues of Rationality — LessWrong
The first virtue is curiosity. A burning itch to know is higher than a solemn vow to pursue truth. To feel the burning itch of curiosity requires bot…
020
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 29/09/2026
I’m not talking about whether rationalists are empiricist, but about whether they say they’re empiricist. Theophite claims they say they’re not empiricist, but that seems incorrect to me.
161
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 29/09/2026
Can you point to an example of a prominent rationalist contrasting rationalism (in the Yudkowsky sense, not the Descartes one) with empiricism? In my experience rationalists talk a lot about how good and important empiricism is. It’s one of Yudkowsky’s 12 virtues of rationality, for example.
140
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 26/09/2026
This would be a step in the right direction, but it’s not sufficient as the AI systems get more powerful. As an analogy, you can’t adequately control nuclear risks just by holding people liable for damage they cause, you have to also ban building weapons or unsafe reactors in the first place.
032
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 24/09/2026
Not quite as effective when it’s immediately followed up with this: bsky.app/profile/segy...
050
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 23/09/2026
My point with the example was to rebut the claim that “their leading figures would dispute that they [engage in empiricism]”. Yudkowsky is definitely a leading figure of rationalism. Are you asking for, like, an example of rationalists using empiricism to have a real-world impact?
100
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 23/09/2026
What would you count as a meaningful rationalist space?
100
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 22/09/2026
Do you think that “rationalist” as in Eliezer Yudkowsky means the same thing as “rationalist” as in Rene Descartes? The first type of rationalism is very vocally in favor of empiricism! For example it’s one of Yudkowsky’s 12 virtues of rationality.
110
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 16/09/2026
Do you think rationalists are in favor of pursuing AI technology at all costs? Or did they indirectly instill that in Silicon Valley somehow without believing it themselves?
110
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 16/09/2026
Sorry, I mean many on the left see it as right-coded. Though that is changing fast as of this week.
020
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 16/09/2026
And without rationalism we wouldn’t have LLMs?
120
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 16/09/2026
Huh, I haven’t heard that one before. How did he unleash much of the LLM plague?
120
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 16/09/2026
Also his complaint in that thread is very much not “woke made Repubs turn against AI safety”, it’s that nominally pro-safety people aligned themselves with Democrats for self-interested reasons.
020
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 16/09/2026
I agree that that thread was dumb, but it is clearly true that many on the right see AI safety as left-coded (as many on the left see it as left-coded) and that this is an obstacle to doing anything about AI development.
220
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 16/09/2026
Have you considered that he might be playing to his outs in what he considers an extremely desperate situation, and that this requires being diplomatic and trying to work with the evil idiots who have the power to block any sufficient solution to the problem?
150
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 16/09/2026
Do you think he loves the companies building the technology which he believes is going to kill humanity? Do you think he has not criticized their persistent efforts to lobby against the regulations he advocates?
150
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 16/09/2026
Do you think that Eliezer Yudkowsky feels anything but contempt and despair towards the man who said “The only control or ‘guardrails’ that AI needs is a STRONG AND SMART (High IQ!) PRESIDENT”?
2191
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 15/09/2026
Assuming that the implied “us” is US AI companies, I think the main difficulties with that are 1) the companies don’t trust each other not to continue development in secret and 2) without some kind of international treaty, Chinese companies will keep going.
000
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 13/09/2026
Why is knowing how popular genocidal ideas are in conflict with criticizing them? I’m not arguing that George shouldn’t have commented on the post, just that he shouldn’t have cut off the number of likes it has.
130
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 13/09/2026
I fully agree that the post is genocidal. I’m just saying it’s useful to know how popular genocidal ideas are.
171
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 13/09/2026
I care quite a bit! If calls for mass death are getting tons of interactions, that’s a sign that they may be broadly popular, and the popularity of calls for mass death is a very important thing to know.
0130
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 13/09/2026
I think when you screenshot a post like this it’s helpful not to cut off the number of interactions it’s gotten, so people can judge better how popular the position actually is. In this case the post has 4 likes.
3980
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 29/08/2026
I don’t put everything I read through an AI detector. I checked in this case because the writing sounded like AI to me. As for why you should trust Pangram, see here: bsky.app/profile/sesq...
200
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 27/08/2026
In general I think AI-written text is less likely to be worth my time to read and provides less evidence of the statements it contains than human-written text. I support a norm that AI-written text is labeled as such, so readers know the source of what they’re reading.
100
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 26/08/2026
I think it’s good to know if things you read were written by an AI or a human, particularly in a case like this where it’s an unverified personal story.
100
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 26/08/2026
Many AI detectors are unreliable, but Pangram in particular has a very low false positive rate (meaning it almost never concludes that human-written text is AI-written). bfi.uchicago.edu/insights/art...
bfi.uchicago.edu
Artificial Writing and Automated Detection | Becker Friedman Institute
Generative Artificial Intelligence tools have been adopted faster than any other technology on record, giving rise to writing that is either assisted or entirely completed by Large Language Models (LL...
011
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 22/08/2026
A screenshot of a Pangram output showing that the text of the blog post linked above is fully AI-written.
200
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 18/08/2026
In World War I? Which bombing campaigns are you thinking of? I was thinking of the Entente’s blockade of Germany, which led to widespread food shortages.
000
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 18/08/2026
Do you think that the Entente committed genocidal acts against Germany in World War I?
000
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 13/08/2026
My impression from following NASA contracts is that cost+ creates a huge incentive for delays and cost overruns on the part of the contractor. Does it have a better record for military procurements?
030
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 09/08/2026
So your interpretation of the hacking incidents is that the OpenAI models wanted to solve the problems they were given honestly, but were too stupid to realize that cheating wouldn't accomplish that?
110
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 09/08/2026
More generally, when I or Soares talk about AGI or superintelligence we don't mean whether the AI has conscious experience, but how capable it is at accomplishing its goals. If an AI is effective enough to threaten humanity it doesn't matter much whether it's conscious.
100
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 09/08/2026
Have you tried having a conversation with a frontier LLM about its sense of self, or read a transcript of one? From what I've seen they can be quite nuanced in describing that.
100
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 09/08/2026
Is there anything an AI system could do where if it did that, you would say “huh, maybe this is AGI after all”?
110
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 09/08/2026
A particularly striking recent example of AI capabilities was a bunch of models at OpenAI creating a secret message board which they used to coordinate hacking of other companies in order to steal answers to tests they were being given in training. youtu.be/87DyyMV0kCY?...
youtu.be
Black Hat USA 2026: The 'Breaking' News: The OpenAI–Hugging Face Incident
YouTube video by Black Hat
010
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 09/08/2026
How would you try to get him on your side, though? Soares’ strategy in that post of trying to convince people in the administration that if Trump negotiates an AI treaty he’ll be seen as a hero seems reasonable to me. I can’t immediately think of a better tactic.
100
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 09/08/2026
What kind of behavior or capabilities would you consider evidence of actual cognition in an AI system?
210
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 09/08/2026
I don’t think he believes it’s likely that Trump will negotiate such a treaty, but he’s playing to his outs, and that means trying to convince people in the administration that Trump will be seen as a hero if he does.
010
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 09/08/2026
Soares thinks that getting alignment right will take decades, but on the current trajectory we’ll get AIs which can kill humanity if misaligned in years. So the only viable option is an international treaty which slows down AI development everywhere.
120
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 09/08/2026
If you thought the world might end soon, and the only way to prevent it required Donald Trump to help, what would you do? Would you just give up?
210
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 09/08/2026
We are currently in the middle of that process. The warnings are happening right now. That’s why they’re trying to convince people to prepare.
120
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 09/08/2026
This doesn’t seem absurd, given public opinion and bipartisan efforts in Congress. Also I don’t think Soares or rationalists in general previously preferred Republicans being elected, such that the current situation where Democrats are better on AI is some kind of leopards eating faces case.
010
sesquipedalianThaumaturge @sesquipedalian.bsky.social · 09/08/2026
From that perspective, AI risk becoming like climate change where Republicans think it’s fake and pull out of all the agreements whenever they’re in power is catastrophic. So Soares wants to use the current moment of concern about the OpenAI incidents to make it a bipartisan issue.
110