Ya, super interesting research area the authors explored of basically trying to answer the question: "Is there a canonical/intrinsic way that concepts/representations/information are 'stored' in the universe/reality?".
They tested that by performing "spectral analysis of over 1100 models - including 500 Mistral-7B LoRAs, 500 Vision Transformers, and 50 LLaMA-8B models ... by applying spectral decomposition techniques to the weight matrices of various architectures", and concluding that "deep neural networks trained across diverse tasks exhibit remarkably similar low-dimensional parametric subspaces", showing that "neural networks systematically converge to shared spectral subspaces regardless of initialization, task, or domain".
Not just philosophically interesting but also has practical implications for being smarter about how to reuse models, model merging, developing more sustainable training and inference algos, etc.
As mentioned in the comments, this is essentially Nick Land's vision, which he has been developing for over 30 years. If anyone is interested in exploring this topic in depth, I have a research project on it: https://retrochronic.com (The Capital Autonomization section of the introduction is particularly relevant.)
Yes. Here's one relevant quote that touches on autonomization and secession:
> Most simply, there is the utilitarian order, in which capital establishes itself as the competitively-superior solution to prior purposes (production of human use-values), and the intelligenic order in which it accomplishes its self-escalation (mechanization, autonomization, and ultimately secession).
Nick Land (2014). Freedoom (Prelude-1a) in Xenosystems Blog. Retrieved from github.com/cyborg-nomade/reignition
> At the Great Midnight at the century's end, signifying culture will flip over into a number-based counterculture, retroprocessing the last 100 years. Whether global disaster ensues or not, Y2K is a singularity for cybernetic culture. It's time to get Y2K positive.
Nietzsche argued that genius is more frequent than we think, but that something else is missing for its realization ("the five hundred hands"):
> In the realm of genius, might the “Raphael without hands” — the term understood in its broadest sense — be not the exception, but the rule? — Genius is perhaps not so rare after all: but the five hundred hands it needs to tyrannize the καιρὁς, “the right time” — to seize chance by the scruff of the neck! [0]
The interesting thing is that, for the "Father of Accelerationism" (Nick Land), AI Doomerism (doom for humans, at least for human identity) and Accelerationism (which for Land is just another label for capitalism: 'The label "accelerationism" exists because "capitalismism" would be too awkward.'[0]) are not opposed at all. And capitalism does not need to get elected.
(Land follows the above quote with "(But the reflexivity of the latter [capitalismism] is implicit.)"[0], which specifies that, for Land, more precisely, "Accelerationism is simply the self-awareness of capitalism"[1].)
[0] Nick Land (2018). Outsideness: 2013-2023, Noumena Institute, p. 71.
[1] Nick Land (2017). A Quick-and-Dirty Introduction to Accelerationism in Jacobite Magazine. Retrieved from github.com/cyborg-nomade/reignition
i don't know, to me they are very different things - accelerationists might be really calling for Better Capitalismism, but that's only because chatbots (the thing you are accelerating) are really good at math, and math is important for making money. if it weren't good at making money, literally nobody would care, kids would not be CS and math majors, they wouldn't care about STEM. they only care because $. But most real problems, including human problems, are not math problems.
this is a huge blind spot in the whole, rationalist and broader STEM cultural-professional community: math isn't the best way to solve problems, most problems are not math problems. SOME of school might be math problems, and it feels good to be a Doctor or a Software Developer Engineer and get your kids to practice "problem solving" - no, they are practicing math problems, not problem solving.
for example there's no math answer to whether or not a piece of land should be a parking lot, or an apartment building, or a homeless shelter, or... you can see how just saying, "whoever is the highest bidder" - that's the math answer, that's why capitalism and accelerationism are related to their core - isn't a good answer. it pretends to be the dominant way we organize land, and of course, it isn't the dominant way we organize land usage anywhere at all, even if we pretend it is. there's no "bidding" for whether a curb should be a disabled parking spot, or a bike lane, or parking, or a restaurant seating, or a parklet, or... these are aesthetic, cultural choices, with meaningless economic tradeoffs. it's not about money, so it's not about math, so math does not provide an answer. there are lots of essential human questions that cannot even be market priced, such as, what should we pay to invent new cures to congenital, terminal illness in children? parents, and a lot of people, would pay "any" price, which is a market failure - but there are a lot of useful political answers to that question. a chatbot cannot answer that question, and it would struggle to take leadership and get elected to answer that question.
mathematicians are basically never elected. so chatbots would not be. and elezier yudlowsky would not be. are you getting it? capitalism does definitely need to be elected, you might think it wins every election but it very often loses at the local level!
i am agreeing with Hashem Sarkis dean of the MIT SAP and kind of disagreeing with Bong Joon-Ho, for further reading.
Expected outcome. Nick Land and the CCRU have explored how capitalism operationalizes science fiction (distilled in the concept of Hyperstition). Viewed through this lens, prices encode "distributed SF narratives." [0]
[0] Nick Land (1995). No Future in Fanged Noumena: Collected Writings 1987-2007, Urbanomic, p. 396.
reply