数式日

圏論を理解したい人

[数式日 @suushikijitsu]

@airkatakana but can you solve it by uploading mp3s of eye of the tiger and 90s shonen openings

[Harlan Stewart @HumanHarlan]

Um actually they asked the AI to maximize paperclips. It’s just doing what it was told

[数式日 @suushikijitsu]

@ebarenholtz Chalmers had a paper on this https://t.co/PCv8LD2E1E
but I'm not sure what development there's been since this paper.

[数式日 @suushikijitsu]

@EyeBDru @g_shullenberger the argument here is that it will replace ALL people and ALL work whether it is low-quality or high-quality, bitchy or enthusiastic, etc.

[数式日 @suushikijitsu]

@OttokarHochman Can we have an even further top-right with, like, the Phenomenology of Spirit

[数式日 @suushikijitsu]

@woke8yearold @softminus @teortaxesTex Giving the constitution O(n^2) space complexity in the number of leadership generations, à la The 12 Days of Christmas. Also each one is the "crystallization of the collective wisdom of the Communist Party of China" during that period, which reminds me of FF7's Materia

[数式日 @suushikijitsu]

@woke8yearold @softminus @teortaxesTex The other thing I like is that the CPC constitution not only spends a paragraph on the ideological residue of each leadership generation, but during that paragraph calls out its connection to each previous generation's residue

[数式日 @suushikijitsu]

@OttokarHochman

[yoshimi battles video game addiction @nise_yoshimi]

PKD really nailed something about the Nazis here I feel. From the man in the high castle

[数式日 @suushikijitsu]

@fakepalindromes would this be equally true for women? like it's not the organ where the work actually happens but same is true for men (just there's less work)

[数式日 @suushikijitsu]

@OttokarHochman I'm still pretty into calligraphy so I guess my continued insistence on "hand"-coding some stuff gels

[数式日 @suushikijitsu]

@OttokarHochman @deevrod think it was fairly un-mainstream/subcultural in the US even in the 00s tho? just, subcultures you and I were part of/adjacent to maybe

[数式日 @suushikijitsu]

@AndyMasley @ostalugo Yeah the average American isn't that big of a contract bridge enthusiast

[数式日 @suushikijitsu]

@sukikage_ I'm going to read the first 3 chapters

[Aminadad Mercilesse Butcher @OttokarHochman]

@3mericanj0hnson aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa

(This is all from Carl Schmitt's Glossarium)

[Holly ⏸️ Elmore @ilex_ulmus]

Okay finally looking of AI 2040 and I'm going to repeat what many others have said: "passing the torch to AI" is not an acceptable goal. Like AI 2027, it shows sobriety about the risks but still an odd partiality to AI futures over human futures. And some magical thinking about negotiating with AI agents vs human agents. Frankly, I believe author Ryan Greenblatt and perhaps others really really really want an AI future, and that bias is criminal in a forecast bc it easily hides behind the subjective component of expert judgment.

There is a bias in the technical AI Safety world to think that technical problems are solvable and coordination problems are impossible. And I think that covers why the AI 2040 prefers plan A (hand the torch to AIs later) over plan S. (shut it all down).

Idk if they get to a principled reason why doing ANY further AI research means eventually giving up control, but I don't see one. Feels like there ought to be something between Plan S and Plan A-- that's what I imagine happening during a Pause: research to figure out how far we can safely (and with consent) go, while retaining the option to shut it all down by (among other enforcement things) not advancing the frontier.

Especially if the reason not to do plan S is thinking that cooperation between human parties will eventually break down, feels like they aren't fairly considering the risk that loss of control to AI puts us at FOREVER. There's no easy path here as far as cooperation. I wouldn't give up on other humans so quickly. In Plan A, if cooperation between humans and the machines breaks down at any point after 2040, we're dead!

Also there's a bias towards what feels like "elegance" but is actually just handwaving in the AI 2027/2040 worlds that I think easily becomes personal conflict avoidance. They fear the AI company lobbies in a way I think they wouldn't if they didn't live in San Francisco surrounded by AI industry friends and loved ones. I think personal avoidance bleeds into their judgment here. I'm just getting started, but it doesn't feel like they really considered what a liability it is not to challenge the power of the AI lobby as soon as possible. Like you can't really trust plan A to proceed according to plan if the AI lobby is not reined in.

Below you can see they make a big deal about plan S cooperation between humans eventually breaking down, basically saying this is fatal, but then elide over whole huge unsolved problems like how to fucking build aligned AI under plan A.

[Aminadad Mercilesse Butcher @OttokarHochman]

Underrated weird feature of Comte is that while he rejected Saint-Simonian socialism as *politics* he actually basically retains a Saint-Simonian *theory of change*: his envisioned social base for positivism is an alliance of "philosophers, proletaries, and women", and he...

[数式日 @suushikijitsu]

@gossip_girl_10x @BIMBOSATTVA_ one wonders why not just give an age/birthyear range since it's not really any more tokens.
maybe generations are our imperial era names

[数式日 @suushikijitsu]

@justjoshinyou13 @AndyMasley We just need fire, metal and wood tribes