corbin @ corbin @awful.systems

Posts

19

Comments

246

Joined

2 yr. ago

He's too young to grok why operating systems exist. (So am I...) To be fair, we're outgrowing the concept in several areas; folks say that Web browsers, unikernels, blockchains, and now apparently LLMs are all replacements for the operating system, because they resemble each other in functionality sometimes.

I would hate to be his child.

What a coward, only spouting death-threat rhetoric on point-to-point lines and not in public. Presumably he understands that his opinions are vile, and understands that the public would thrash him until he can no longer hold those opinions, but doesn't understand that this means that his attitude needs to be adjusted.

If he were a grifter, yeah. However, he's a true believer, so instead he rants about how modern machine-learning methods are dangerous and could escape at any moment; he genuinely believes that Skynet is being born inside some cloud right now.

Let me know @corbin@defcon.social if you actually get LLMs to produce useful code locally. I've done maybe four or five experiments and they've all been grand disappointments. This is probably because I'm not asking questions easily answered by Stack Overflow or existing GitHub projects; LLMs can really only model the trite, not the novel.

Indeed, this is also the case for anything packaged with #Nix; we have over 99% reproducibility and are not planning on giving that up. Also, Nix forbids network access during compilation; there will be no clandestine queries to OpenAI.

Yud tried to describe a compiler, but ended up with a tulpa. I wonder why that keeps happening~

Yud would be horrified to learn about INTERCAL (WP, Esolangs), which has required syntax for politely asking the compiler to accept input. The compiler is expressly permitted to refuse inputs for being impolite or excessively polite.

I will not blame anybody for giving up on reading this wall of text. I had to try maybe four or five times, fighting the cringe. Most unrealistic part is having the TA know any better than the student. Yud is completely lacking in the light-hearted brevity that makes this sort of Broccoli Man & Panda Woman rant bearable.

I can somewhat sympathize, in the sense that there are currently multiple frameworks where Python code is intermixed with magic comments which are replaced with more code by ChatGPT during a compilation step. However, this is clearly a party trick which lacks the sheer reproducibility and predictability required for programming.

Y'know, I'll take his implicit wager. I bet that, in 2027, the typical CS student will still be taught with languages whose reference implementations use either:

1. the classic 1970s-style workflow of parsing, tree transformation, and instruction selection; or
2. the classic 1980s-style workflow of parsing, bytecode generation, and JIT.

It's the combination of big imaginations and little real-world experience. In Friendship is Optimal, the AGI goes from asking for more CPUs to asking for information on how to manufacture its own CPUs, somehow without involving the acquisition of silicon crystals or ASML hardware along the way. Rationalist writers imagine that AGI will somehow provide its own bounty of input resources, rather than participating in the existing resource economy.

In reality, none of our robots have demonstrated the sheer instrumentality required to even approach this sort of doomsday scenario. I think rationalists have a bit of the capitalist blind spot here, imagining that everything and everybody (and everypony!) is a resource.

Going a bit on side venture here with this comment, but I'd so much prefer this to be well written up instead of a two and half hour video. And it's so ubiquitous, watching videos over people reading, and I can't help but think it's extremely hurtful on a cultural/societal level.

Side sneer: Nah, video is great. Watching an hour-long video is a standard part of research for the working software engineer. Lots of content from conferences like Strange Loop, DEFCON / Black Hat, and CCC is only available as video, and was intended to be watched or accompanied by slides.

...But I may be biased, as I have no problem performing 2hr speedruns. Lately I've been running games with 3hr or 4hr routes, and I'll readily agree that it's not for everybody.

To be honest, I think it would be hilarious if Texas were annexed by Mexico, even just a little bit. Texans deserve a competent government who will invest in them and conduct reasonable diplomacy, and they're not getting that from Texas.

I could just as easily say that this is a fundamental design flaw shared by Bluesky and Discord; e.g. Signal and IRC don't have this problem. Security isn't just about response to criticism, but about making design choices which protect users.

NSFW, because this is a pattern for him. So, he used to work on Beaker, a Web browser built on top of the Dat/Hypercore DHT. I recall my experience chatting with him and others on IRC. I had been interested because Dat and Beaker were supposedly built with ocap theory, and at the time I was helping to produce a capability-safe object-oriented programming language. Relevant highlights:

• He did not grok the idea that users might dig below the chrome and directly access APIs. This dovetailed with a lackluster approach to security. In capability theory, users are expressly permitted to do anything they are capable of doing; but Beaker's philosophy was that users ought to restrict themselves to only clicking buttons in Beaker's chrome.
• In general, interoperability was not a big priority. I'm not sure if there's multiple Hypercore implementations yet, but at the time, there was only one reference implementation and not enough documentation to reimplement it from scratch. So, I wouldn't be able to federate with their DHT using my custom software.
• I didn't know who the project leaders were. One time, one of the project leaders came onto IRC, and I made the mistake of greeting them. As a result, I was immediately banned from their IRC channel. However, none of them knew how IRC works, and so they did not kick me; in the aftermath, I listened as they went around the room and disavowed me, covering their asses by explaining that they didn't know who I was or why I was in the room.

Those first two points rhyme with his actions here. The third point is where I think we can see things heading in the future.

It's somewhat infuriating that all of my datacenter adventures have ended up with lacerations on my hands at a minimum, and yet he did not have a single object fall on him. It's almost enough to make me believe in a hateful sadistic deity who protects billionaires.

Without disclosing anything specific from my time at Google, I want to emphasize that he's a narcissist. Most of his posts were dripping with delusions about himself, the industry, the nature of software engineering, and the metaphysics of computing. It is a minor surprise that he is not a cult leader like Yudkowsky or a right-wing talking head like Damore. From the inside, Googlers might say "he lacks impact;" out here, we say that he lacks focus.

NSFW: this article, mentioned but not linked by the Substack author, is good reading if you want to know about the inciting drama. Choice quote:

Personally I always default to dismissing the chuddy “you radicalized me” explanation as a manipulation tactic coming from people who were already wanting to go there and just looking for an excuse. Also adjacent to abuser logic, like “look what you made me do.” I don’t buy that nazi furry radicalizing, of all things, happens to neutral people.

They're cryptofascists. Here is my favorite example; note that he replies to everybody else but not directly to me, because he is a coward.

If you ever happen to chat in-person with this sort of highly-concerned moderate, feel free to grill them about how they would deal with violent fascists. Either they cave, or they'll eventually conclude that their immense powers of rhetoric allow them to verbally defuse Nazis somehow. In this latter case, point out that they can't even convince you that punching Nazis is wrong, and conclude that they must not be very good at rhetoric.

Sorry, but I can't even sneer properly at this sort of cowardice. It's pathetic to the point where ridicule is the only response I can emotionally justify.

At risk of going NSFW, it's obvious that none of these folks have read Singer 1971, which is the paper that kickstarted the EA movement. This paper's argument has a massive fucking hole right in the middle.

Without cracking open the paper, I seem to recall that it is specifically about Oxfam and famine in Africa. The central claim of the paper is that everybody should donate to Oxfam. However, if one is an employee of Oxfam, then suddenly the utilitarian arithmetic fails; his argument only allows for money going from non-Oxfam taxpayers to Oxfam employees.

Can't help but notice how the main problem with EA charities is the fucking nepotism. Almost as if the EA movement rests on a philosophical foundation of ignoring when charities employ friends of donors.

Nah, they're okay with it because it reinforces their belief that a person is either high-empathy or low-empathy, with higher EQ being better. In general, conservatives love standardized tests and grades, because it grants the appearance of merit, which is essential for meritocracy.

Unlucky 10000: There is an EQ, or emotional quotient, and I was given an EQ test in high school (like age 17-18, don't remember exactly). Fortunately, it was just done for fun by a lone teacher, but I could see it becoming popular in a future school system.

Don't forget its other use: corralling high-IQ children into Talented & Gifted programs. Gotta stigmatize them early. (It's okay, I'm allowed to joke about this; I maxed out an IQ test as a child and was shoved into T&G for grade school.)