nk

If you're in for an hour long rant, here a motivational for Neural Network Field Theory. The video also has a short overview of Neural Network Gaussian Processes and Quantum Field Theory. Includes a tangent on Neural Tangent Kernel theory. #NNGP #NNFT #QFT youtu.be/ZSmORp3Bm2c

@backslashvarphi 45% of the US stock market pivoted to AI so that we can finally overcome the perpetual plot.plot() to ax.plot() refactor

a large % of physicists only use claude to change the color on plots or whatever and refuse to use codex because it's unintellectual and spiritually unpure

@millerman we're levitating

@subcountability Yes

I bought more books than I can read in a lifetime because I'd like the library to be a resource and inspiration not only for myself but for family, friends, colleagues, and partners

@kirancodes One of the longest online ML tools, detexify from 2009, let's you find symbols by drawing them in a paint like style detexify.kirelabs.org

never writing latex by hand ever again. You are not ready.

@subcountability Very broadly the racist part of twitter/the internet

Google translate recognizes “gibs”, which I assumed was a word only we used.

I'll gonna also disagree with the mouse on this, for similar reason as I don't think the want for transfinite induction/recursive is an argument for a cardinal hierarchy. Yes for the set of reals ℝ to exists we need something of size |2ᴺ|, but we do not need the set of reals to do analysis. You posted about reverse math two weeks ago, and there 2-leveled arithmetic proves Bolzano-Weierstrass and continuous real functions Riemann integrable without stipulating that sets beyond ℕ exists. Say you code the ᵢ'th rational as (-1)ˢⁱ · nᵢ / dᵢ the key-value pair as (i, (sᵢ, nᵢ, dᵢ)) and then a real is just a subset of ℕ⁴, say coded as 3ⁱ · 5ˢⁱ · 7ⁿⁱ · 11ᵈⁱ Set theory may enable you to drop the desired equivalence relation of Cauchy rationals with extensional equality of sets, but that is a convenience and not necessary to do math. And we can push coding further - we can have weaker notions than the set of natural numbers and still do a lot of analysis. Ofc it gets more an more inconvenient. But suffice to say Brouwer did a lot with spreads and computable functions. No real "reliance" on the naturals as completed infinity (and not even LEM.) As for the powerset-discussion: On the other hand, we can equip first-order with all finite types, i.e. iterate function spaces and make functions object, and thus get a convenient theory without these: Ofc, at this point you're halfway into type theory anyway. Just to be sure, I personally want functions spaces like 2ᴺ and so on (And with a weaker logic this doesn't mean you got "powersets" for everything, but let's not go there here ;P) I'm not against the cumulative hierarchy by any means, but I don't quite know if "we" "need" it. I can't tell if Neumann ordinals are technically necessary for the early descriptive set theory - but of course I don't want to make my live inconvenient either. And I don't only come from a computation-for-physics angle, which I espouse to myself when I think of things. As I said in this thread Z ⊬ ∃λ. (λ = ω + ω) while V_{ω+ω} ⊨ Z and so all the nice topos of set properties kick in without deep sets. If we buy into the idea that such topoi provide what "we" want from math, then it means we don't need to go deep (in rank). I would go as far as saying nobody "needs" ω³ being a Neumann ordinal - which on the other hand is not at all to say that I don't think there should be droves of students and researchers exploring the hierarchies. That's good too. cc @iz_dmr x.com/mouse_math/sta…

@mouse_math Which style do you enjoy? I'm of many minds. I have "provided" a lot of info on Wikipedia over the years, and often not leading to nice articles, but the resource aspect for me outweighs other aspects. There's also lots of older students lost stuff, e.g. en.wikipedia.org/wiki/Ordinal_a…

"This article is written like a textbook. Please help rewrite it in a neutral, encyclopedic style."

despite everything

@y2k_mischief he just isn't well understood by the general populous; maybe will never be

sick of mischief's snide comments

I learned this year that vanilla NF has ℕ and 𝒫(ℕ) and {s ∈ 𝒫(ℕ) : |s|=1} and also functions from ℕ to coded lists of naturals, but the "bridging" function like isn't automatically there, due to the stratification mechanics. Existence of bijection witnessing |X| = |{{x} : x ∈ X}|, even for X=ℕ, is not implying the there much stronger existence of the particular bijection x ⟼ {x}. It's related to the old Frege definition of naturals being treated as equivalence classes, not like the Neumann lego blocks.

Finitists to the left of me, New Foundations to the right Here I am Stuck in the middle with ZFC augmented by large cardinal hypotheses

@robertskmiles syc up with x.com/artchad/status…

There's a big "Stop the AI Race" March in SF right now

@tv4965109947686 I approve.

これで超幾何関数、勉強させていただく。

@ADELDERVISION ?

@subcountability Lambda, Bextor, Grindr. What am I looking at.

@GFreiNews Finanzmensch heranzüchten? Auto-Fetischismus start Kunstgeschichte? Na. Deine Negativ-Beispiele sind nicht überzeugend, scheint alles wichtig für Bildung & Latein wird eh diskutiert. Aber Fazit is wohl, dass einfach nicht genüg Zeit is; oder eher, dass die Schüler schwach sind.

Kann mir mal jemand unser Schulsystem erklären? In der Schule werden Dinge gelernt, die absolutes Spezialwissen sind und die hinterher nur jeder Hundertste oder gar Tausendste benötigt. Musiknoten, analytische Geometrie, Kunstgeschichte, chemische Gleichungen, Latein – seien wir mal ehrlich: 80 % unseres Schulwissens war nutzlos und wurden vergessen, weil wir es nie wieder benötigt haben. Aber nichts von den Dingen, die wirklich absolut jeder braucht, wird in der Schule gelehrt. Der Autoführerschein, wie Versicherungen funktionieren, Zinsen, Steuererklärung, Geldanlage, Erste Hilfe usw. Warum wird die Zeit unserer Schüler mit Themen verschwendet, die für die meisten komplett nutzlos sind? Und all das, was absolut jeder benötigt, wird uns in der Schule nicht beigebracht?

🇫🇷🇪🇸 🏆 🏴󠁧󠁢󠁥󠁮󠁧󠁿🇦🇷

@LambdaNrx Ich hab auch paar lokalen models probiert (via 'LM Studio' zum downloaden). Auf dem M4 MacBook sind 8 GB varienten zZ. circa so schnell wie in dem screenrecording. Ich rate, dass wir ein, zwei Jahren gute pur lokale coding agents haben. Laptop Dämon ja, aber weniger Glasmensch.

@LambdaNrx Not for long. Is ja eh klar...

Ich würde gerne ein Video machen über KI, Arbeitsplätze, Bullshitjobs und die Zukunft der Wirtschaft. (Hauptthema: wird KI massenweise Arbeitsplätze zerstören und was sind die Konsequenzen?) Will jemand von euch Gast sein oder kennt jemanden, der geeignet ist? Bevorzugt Ökonom.