Hacker News: zyklu5

New comment by zyklu5 in "Embarrassingly simple self-distillation improves code generation"

zyklu5 — Sat, 04 Apr 2026 19:41:25 +0000

Their explanation for why their idea (SSD) might work - precision-exploration conflict hypothesis - is something adaptive decoding also tries to solve.

https://ai.meta.com/research/publications/adaptive-decoding-...

New comment by zyklu5 in "Michael Faraday: Scientist and Nonconformist (1996)"

zyklu5 — Sun, 15 Mar 2026 15:46:47 +0000

Thank you very much for the link.

This letter from Maxwell to Faraday: https://epsilon.ac.uk/view/faraday/letters/Faraday3354 is very surprising to me. Already they are thinking of gravity in field theoretic terms. Just to set context: Riemann's Habilitationsschrift was just 3 years earlier but not to be published till late 1860s.

New comment by zyklu5 in "Manuscript of Ismail al-Jazarī's Ingenious Mechanical Devices (ca. 17th century)"

zyklu5 — Tue, 29 Apr 2025 11:15:52 +0000

Indeed. In fact, it is one of the most amusing aspect of the anglophone west (at least for the last few decades). Despite public perception (by public I mean those who have been to university since the 90s), Western historians of science and mathematics in general have never not acknowledged the previous works of the Persianate civilizations commensurate to their knowledge of them in their time. But somehow in the last few decades professional historians have had to waste time figuratively looking over their shoulders lest they be percieved as being Eurocentric. And, if they were to somehow find a way to show -- requiring whatever hermeneutical gymnastics -- that a prominent scientist was influenced (or even better, had stolen) from some other "cultures" than nothing better! (ex: Copernicus from the Maragha school as an example of interpretive gymnastics)

But, of course, this is one of the symptoms of the degeneration that now afflicts your particular civilization and is bringing about it's inevitable transformation to something else -- but better this than the fate of the Abassids or the Sung.

New comment by zyklu5 in "Electron Band Structure in Germanium, My Ass"

zyklu5 — Tue, 01 Apr 2025 15:48:58 +0000

On the contrary, what is presented by the OP is one of the many reasons that worship of science's heroes, unfashionable for decades, a whiggish pablum, is justified. If great results were birthed fully-formed -- a view I've frankly never heard anyone profess who has bothered to consider such things even briefly -- they would hardly be any heroes. Even little children who reflexively chomp on every superhero film aeroplaned towards their face understand this.

New comment by zyklu5 in "Lemma for the Fundamental Theorem of Galois Theory"

zyklu5 — Mon, 17 Mar 2025 17:01:09 +0000

If I'm being honest I really don't have a good recommendation for a text in intro abstract algebra. I learned it from Michael Artin's Algebra. Artin is a true master -- along with David Mumford, he was the main apostle for Grothendieck style AG in the US -- but his book was not very easy to learn from.

New comment by zyklu5 in "Lemma for the Fundamental Theorem of Galois Theory"

zyklu5 — Sat, 15 Mar 2025 18:55:43 +0000

There is indeed a deep connection between what is going on behind Arnold's proof and the classical Galois theory. But it needs quite a bit of sophistication to flesh out properly (not apparent in his famous lectures given to high school kids). There is a Galois theory for Riemann surfaces over algebraic functions where the coverings behave like fields do in the classical correspondence. If any one is interested, check out chapter 3 of Khovanskii's Galois Theory, Coverings and Riemann Surfaces.

New comment by zyklu5 in "Lemma for the Fundamental Theorem of Galois Theory"

zyklu5 — Sat, 15 Mar 2025 16:12:51 +0000

Let me take this opportunity to post one of the best texts on Galois Theory I have read -- and I had to go through quite a few while preparing for a class.

https://pages.uoregon.edu/koch/Galois.pdf

The subject is developed very naturally and every idea is beautifully motivated. It begins with a quick one chapter intro of Arnold's proof of Abel-Ruffini.

Richard Koch's home page (https://pages.uoregon.edu/koch/) has other examples of his fantastic pedagogy.

New comment by zyklu5 in "It is as if you were on your phone"

zyklu5 — Mon, 10 Mar 2025 17:32:53 +0000

You know how in every zombie movie there's a bit where our intrepid protagonists must blend in to avoid capture. I think this is that sort of thing.

New comment by zyklu5 in "Mathematics in the 20th century, by Michael Atiyah [pdf] (2002)"

zyklu5 — Sun, 09 Feb 2025 17:05:20 +0000

Atiyah is truly one of the giants of modern mathematics. I remember long ago I struggled through a reading course of his and Bott's Yang-Mills paper in graduate school. Like many great works of math it too had that paradoxical characteristic of transforming seemingly 'non-mathematics' into mathematics* by reversing the usual direction of application of one to the other, in this case, from physics to math. It would start a whole movement that'll produce much of modern geometries greatest hits like Donaldson's (his student) theorem in 4 manifolds to Witten's great papers.

* A reason I think modern LLM architecture as they currently stand with their underlying attention mechanisms will not produce interesting new mathematics. A few other ideas are going to be needed.

New comment by zyklu5 in "Children's arithmetic skills do not transfer between applied and academic math"

zyklu5 — Fri, 07 Feb 2025 21:46:54 +0000

And why should I simply assume that "Education Economists"* really know the subject they purport to talk about? Because they are credentialed members of university departments with some label? Because a few of them won some Bank of Sweden award?

Just because a particular department or field of study exists in academia does not magically give them the imprimatur you think it does.

* Btw, I know for a fact that a few of them are not "education economists"

New comment by zyklu5 in "Notes on OpenAI o3-mini"

zyklu5 — Sat, 01 Feb 2025 18:18:58 +0000

Claude is still better in my opinion.

There's a suite of code-related tasks -- covering a diversity of areas, including dev ops, media manipulation etc., derived from issues I have faced over the years -- I perform for every new release. No model has solved the set of issues solved in one go but Claude still remains the best.

An example of the sort of problems in the suite:

> I have a special problematically encoded mp4 file with a subtle issue (something I ran into a couple of years ago while fixing a bug in a computer vision pipeline). In the question prompt I also pass the output of ffprobe and ask for the ffmpeg command that'll fix it. Only Claude has figured the real underlying issue out (after 4 interactions).

New comment by zyklu5 in "Human corpses keep moving for over a year after death, scientist says (2019)"

zyklu5 — Tue, 28 Jan 2025 16:07:37 +0000

Thank you so much for this! I have read a fair amount of Lem but somehow completely missed this one.

My Lunch with V.I. Arnold [pdf]

zyklu5 — Mon, 20 Jan 2025 16:49:44 +0000

Article URL: https://gomboc.eu/docs/99.pdf

Comments URL: https://news.ycombinator.com/item?id=42770474

Points: 1

# Comments: 0

New comment by zyklu5 in "Rational or not? This basic math question took decades to answer"

zyklu5 — Sat, 11 Jan 2025 13:03:47 +0000

Here is Frank Calegari's excellent talk on this work:

https://www.youtube.com/watch?v=znBdPEyDScY

Mechanistic Interpretability (Of Large Language Models) [video]

zyklu5 — Mon, 23 Dec 2024 18:58:54 +0000

Article URL: https://www.youtube.com/watch?v=UGO_Ehywuxc

Comments URL: https://news.ycombinator.com/item?id=42496736

Points: 1

# Comments: 0

New comment by zyklu5 in "Fermat's Last Theorem – how it’s going"

zyklu5 — Thu, 12 Dec 2024 18:16:15 +0000

Re: that dig at 4-manifolds

Are you aware of the book on [The Disc Embedding Theorem](https://academic.oup.com/book/43693) based on 12 lectures Freedman gave roughly a decade ago.

New comment by zyklu5 in "Everyone is capable of, and can benefit from, mathematical thinking"

zyklu5 — Thu, 21 Nov 2024 19:56:15 +0000

I think you've simply redefined genius. Many years ago I read an article on youth football, if I remember correctly, and in it there was a bit about the writers visit to the Ajax Youth Academy. In it he writes of a moment during practice when a plane flies over and all the 7(?) year olds on the pitch look up to see it except for one kid who keeps his eyes on the ball. That kid (of course) grows up to be a very good midfielder for Real (I'm forgetting the exact details, I think its Wesley Sniejder?). My point is: whatever that motive energy is that manifests as the single minded pre-occupation with math at an age when everybody else's attention is all over the place is that inherent thing that people call genius. I have read many of Thurston's non-mathematical writings about himself and in it this sort of singular pre-occupation is also clear -- which is why he developed his preternatural geometric vision.

New comment by zyklu5 in "Show HN: Tinder, but to Decide What to Eat"

zyklu5 — Mon, 04 Nov 2024 06:43:24 +0000

Maybe I am misinterpreting but this is a beautifully crafted bit of wry humour. :D

New comment by zyklu5 in "Hofstadter on Lisp (1983)"

zyklu5 — Thu, 17 Oct 2024 06:53:01 +0000

You must interact with more interesting people than I because to me LLMs have demonstrated as much "common sense, insight, creativity, self awareness, humor" as the average person I run into (actually maybe more but that makes me sound crazy to myself).

New comment by zyklu5 in "Big advance on simple-sounding math problem was a century in the making"

zyklu5 — Tue, 15 Oct 2024 19:54:11 +0000

As a first pass, check out the pop history book called Fearless Symmetry by Ash and Gross.

Next, and perhaps I shouldn't be suggesting it, Serre's A Course in Arithmetic. Despite it's reputation of terseness it is a great book by one of history's great mathematicians and worth every sweat and tear spent over its short number of pages.

Yet another way is to see the lectures by Richard Borcherds (who won the fields for the moonshine conjecture) on youtube.

Finally, since this hacker news check out William Stein's Modular Forms: A Computational Approach (pdf online)