Ciro Santilli @cirosantilli 40 

ONNX Updated 2025-07-16

The most important thing this project provides appears to be the .onnx file format, which represents ANN models, pre-trained or not.

Deep learning frameworks can then output such .onnx files for interchangeability and serialization.

Some examples:

activatedgeek/LeNet-5 produces a trained .onnx from PyTorch
MLperf v2.1 ResNet can use .onnx as a pre-trained model

The cool thing is that ONNX can then run inference in an uniform manner on a variety of devices without installing the deep learning framework used for. It's a bit like having a kind of portable executable. Neat.

One page to rule them all Updated 2025-07-16

It is true that one image is worth a thousand words, but unfortunately it is also true that one image takes up at least as much bytes as a thousand words!

Having one single page to rule them all is of course the ideal setup for a website, as you can Ctrl + F one ToC and quickly find what you want.

And, with Linux Kernel Module Cheat Ciro noticed that it is very hard to write so much intelligent prose that becomes larger than reasonable to load on a single webpage.

He then started using this technique for everything he writes, including this page and Chinese government.

However, if there are too many images on the page, the loading of the last images would take forever in case users want to view the last sections.

There are two solutions to that:

be traditional and create separate web pages
be bold and load images as they appear on the viewport: stackoverflow.com/questions/2321907/how-do-you-make-images-load-only-when-they-are-in-the-viewport/57389607#57389607
Edit: OK, it was standardized with loading=lazy, without need JavaScript!
Now the last awesome thing would be a method that loads first images in viewport, then those below, and then those above, that would be the ultimate solution.
This question comes close: stackoverflow.com/questions/7906348/change-loading-order-of-images-already-on-page

Ciro is still deciding between those two. The traditional approach works for sure but loses the one page to rule them all benefits.

The innovative approach will work for interactive viewing, but archive.org will fail to load the images for example, and there may be other unforseen consequences.

Wikimedia Commons is awesome and automatically converts and serves smaller versions of images, so always choose the smallest images size needed by the output document. Readers can then find the higher resolution versions by following the page source.

This also comes to mind: motherfuckingwebsite.com

zettelkasten.de/posts/overview/ from zettelkasten:

How many Zettelkästen should I have? The answer is, most likely, only one for the duration of your life. But there are exceptions to this rule.

OpenStax Updated 2025-07-16

These people have good intentions.

The problem is that they don't manage to go critical because there's to way for students to create content, everything is manually curated.

You can't even publicly comment on the textbooks. Or at least Ciro Santilli hasn't found a way to do so. There is just a "submit suggestion" box.

This massive lost opportunity is even shown graphically at: cnx.org/about (archive) where there is a clear separation between:

"authors", who can create content
"students", who can consume content

Maybe this wasn't the case in their legacy website, legacy.cnx.org/content?legacy=true, but not sure, and they are retiring that now.

Thus, OurBigBook.com. License: CC BY! So we could re-use their stuff!

By Rice University.

TODO what are the books written in?

github.com/openstax/openstax-cms Uses Wagtail CMS. So presumaby they just Wagtail's WYSIWYG.
github.com/openstax/os-webview

Richard Baraniuk on open-source learning by TED (2006)

Source.

Ollama Updated 2025-07-16

github.com/jmorganca/ollama

Ollama is a highly automated open source wrapper that makes it very easy to run multiple Open weight LLM models either on CPU or GPU.

Its README alone is of great value, serving as a fantastic list of the most popular Open weight LLM models in existence.

Install with:

curl https://ollama.ai/install.sh | sh

The below was tested on Ollama 0.1.14 from December 2013.

Download llama2 7B and open a prompt:

ollama run llama2

On P14s it runs on CPU and generates a few tokens per second, which is quite usable for a quick interactive play.

As mentioned at github.com/jmorganca/ollama/blob/0174665d0e7dcdd8c60390ab2dd07155ef84eb3f/docs/faq.md the downloads to under /usr/share/ollama/.ollama/models/ and ncdu tells me:

--- /usr/share/ollama ----------------------------------
    3.6 GiB [###########################] /.ollama
    4.0 KiB [                           ]  .bashrc
    4.0 KiB [                           ]  .profile
    4.0 KiB [                           ]  .bash_logout

The file:

/usr/share/ollama/.ollama/models/manifests/hf.co/mlabonne/Meta-Llama-3.1-8B-Instruct-abliterated-GGUF/Q2_K

gives a the exact model name and parameters.

We can also do it non-interactively with:

/bin/time ollama run llama2 'What is quantum field theory?'

which gave me:

0.13user 0.17system 2:06.32elapsed 0%CPU (0avgtext+0avgdata 17280maxresident)k
0inputs+0outputs (0major+2203minor)pagefaults 0swaps

but note that there is a random seed that affects each run by default. ollama-expect is an attempt to make the output deterministic.

Some other quick benchmarks from Amazon EC2 GPU on a g4nd.xlarge instance which had an Nvidia Tesla T4:

0.07user 0.05system 0:16.91elapsed 0%CPU (0avgtext+0avgdata 16896maxresident)k
0inputs+0outputs (0major+1960minor)pagefaults 0swaps

and on Nvidia A10G in an g5.xlarge instance:

0.03user 0.05system 0:09.59elapsed 0%CPU (0avgtext+0avgdata 17312maxresident)k
8inputs+0outputs (1major+1934minor)pagefaults 0swaps

So it's not too bad, a small article in 10s.

It tends to babble quite a lot by default, but eventually decides to stop.

Oil drop experiment Updated 2025-07-16

Clear experiment diagram which explains that the droplet mass determined with Stoke's law:

Quantum Mechanics 4a - Atoms I by ViaScience (2013)

Source.

American Scientific, LLC sells a ready made educational kit for this: www.youtube.com/watch?v=EV3BtoMGA9c

Here's some actual footage of a droplet on a well described more one-off setup:

Video 2.

Millikan's Experiment, Part 2: The Experiment by Phil Furneaux (2017)

Source. From Lancaster University

This American video likely from the 60's shows it with amazing contrast: www.youtube.com/watch?v=_UDT2FcyeA4

Office space design and remote work Updated 2025-07-16

Working remotely is hard if you don't already highly master the software and enterprise systems used.

Also you don't feel people's love as strongly, and usefulness is built on love, see also Steve Jobs's Pixar office space design philosophy.

But please, give workers a small silent office so that we can concentrate instead of a silly open space, and create an internal social network so people can see what others are doing.

Remote working is much better if the majority of the team also does it, otherwise you will get excluded. Maybe after VR...

Obsidian (software) Updated 2025-07-16

Good:

WYSIWYG
Extended-Markdown-based
help.obsidian.md/Getting+started/Sync+your+notes+across+devices they do have a device sync mechanism
it watches the filesystem and if you change anything it gets automatically updated on UI
help.obsidian.md/links#Link+to+a+block+in+a+note you can set (forcibly scoped) IDs to blocks. But it's not exposed on WYSIWYG?

Bad:

forced ID scoping on the tree as usual
no browser-only editor, it's just a local app apparently:
obsidian.md/publish they have a publish function, but you can't see the generated websites with JavaScript turned off. And they charge you 8 dollars / month for that shit. Lol.
- folders are not articles, you can't see them as as page
- HTML export only as third party plugin:
  - github.com/KosmosisDire/obsidian-webpage-export
  - www.reddit.com/r/ObsidianMD/comments/pf3cw6/any_plugin_to_export_an_entire_vault_as_html_pages/
  Possibly in self conflict with them wanting people to pay 8 dollars / month for their hosting? Crazy.
block elements like images and tables cannot have captions?
- forum.obsidian.md/t/adding-caption-to-images/16431
- forum.obsidian.md/t/captions-for-tables-in-markdown/17240
they kind of have synonyms: help.obsidian.md/aliases but does it work on source code?

Some Obsidian digital gardens that would fit well with OurBigBook:

publish.obsidian.md/discretecs/START+HERE

Oak Ridge National Laboratory Updated 2025-07-16

Located in Tennessee in the East of the United States.

The precursor organization to ORNL was called Clinton Engineer Works, where groundbreaking Manhattan Project experiments and nuclear production took place during World War II

Some key experiments carried out there include:

1943: X-10 Graphite Reactor: prototype Plutonium breeder reactor
isotope separation to purify Uranium-235:

Nvidia Updated 2025-07-16

Open source driver/hardware interface specification??? E.g. on Ubuntu, a large part of the nastiest UI breaking bugs Ciro Santilli encountered over the years have been GPU related. Do you think that is a coincidence??? E.g. ubuntu 21.10 does not wake up from suspend.

Linus Torvalds saying "Nvidia Fuck You" (2012)

Source.

Video 2.

How Nvidia Won Graphics Cards by Asianometry (2021)

Source.

Doom was the first killer app of personal computer 3D graphics! As opposed to professional rendering e.g. for CAD as was supported by Silicon Graphics
youtu.be/TRZqE6H-dww?t=694 they bet on Direct3D
youtu.be/TRZqE6H-dww?t=749 they wrote their own drivers. At the time, most drivers were written by the computer manufacturers. That's insane!

Video 3.

How Nvidia Won AI by Asianometry (2022)

Source.

Nutrient Updated 2025-07-16

Numerical method to solve a partial differential equation Updated 2025-07-16

The finite element method is one of the most common ways to solve PDEs in practice.

Nu (letter) Updated 2025-07-16

Why would physicists use a letter such that:

the upper case version looks exactly like an upper case N. At least that is the correct pronunciation/name/historical successor of $ν$ .
the lower case version looks exactly like a lower case V

Why? Why?????????

Nuclear weapon test Updated 2025-07-16

Nuclear weapon Updated 2025-07-16

Uranium vs plutonium: Section "Uranium vs plutonium Quora answer by Ciro Santilli".

Ciro Santilli is mildly obsessed by nuclear reactions, because they are so quirky. How can a little ball destroy a city? How can putting too much of it together produce criticality and kill people like in the Slotin accident or the Tokaimura criticality accident. It is mind blowing really.

More fun nuclear stuff to watch:

Dr. Strangelove (1964)
en.wikipedia.org/wiki/Chernobyl_(miniseries)
The World Of Enrico Fermi by Harvard Project Physics (1970)
Fat Man and Little Boy (1987) shows a possibly reasonably realistic of the history of the development of the Trinity

Tour of a nuclear misile silo from the 60's by Arizona Highways TV (2019)

Source.

Video 2.

The Ultimate Guide to Nuclear Weapons by hypohystericalhistory (2022)

Source. Good overall summary. Some interesting points:

youtu.be/8uIPQBOCJ64?t=2946 talks about the difference between tactical and strategic nuclear weapons
youtu.be/8uIPQBOCJ64?t=3291 mentions variable yield devices, this is the main new thing Ciro Santilli learned from this video
youtu.be/8uIPQBOCJ64?t=3416 discusses if a strategic nuclear weapon usage would inevitably lead to tactical nuclear weapon escalation. It then mentions one case in which a possibly comparable escalation didn't happen: the abstinence of using chemical weapon during World War II.

Nuclear reactor Updated 2025-07-16

Some of the most notable ones:

1942: Chicago Pile-1: the first human-made nuclear chain reaction.
1943: X-10 Graphite Reactor: an intermediate step between the nuclear chain reaction prototype Chicago Pile-1 and the full blown mass production at Hanford site. Located in the Oak Ridge National Laboratory.
1944: B Reactor at the Hanford site produced the plutonium used for Trinity and Fat Man

Not every

x

belongs to the elliptic curve over a non quadratically closed field Updated 2025-07-16

One major difference between the elliptic curve over a finite field or the elliptic curve over the rational numbers the elliptic curve over the real numbers is that not every possible

x

generates a member of the curve.

This is because on the Equation "Definition of the elliptic curves" we see that given an

x

, we calculate

x^{3} + a x + b

, which always produces an element

y^{2}

But then we are not necessarily able to find an

y

for the

y^{2}

, because not all fields are not quadratically closed fields.

For example: with

a = 1

and

b = 1

, taking

x = 1

gives:

y^{2} = 1^{3} + 1 \times 1 + 1 = 3

(1)

and therefore there is no

y \in Q

that satisfies the equation. So

x = 1

is not on the curve if we consider this elliptic curve over the rational numbers.

That

x

would also not belong to Elliptic curve over the finite field

F_{4}

, because doing everything

mod 4

we have:

0 * 0 1 * 1 2 * 2 3 * 3 = 0 = 1 = 4 = 9 = 0 = 1 mod 4 mod 4 mod 4 mod 4

(2)

Therefore, there is no element

y \in F_{4}

such that

y \times y = 2

y \times y = 3

, i.e.

2

and

3

don't have a multiplicative inverse.

For the real numbers, it would work however, because the real numbers are a quadratically closed field, and

3 \in R

For this reason, it is not necessarily trivial to determine the number of elements of an elliptic curve.

North America Updated 2025-07-16

Norm induced by the complex dot product Updated 2025-07-16

Given:

x = \sum_{k = 1}^{n} a_{k} + b_{k} i \in C^{n}, a_{k}, b_{k} \in R

(1)

the norm ends up being:

∣ x ∣ = \sum_{k = 1}^{n} a_{k}^{2} + b_{k}^{2}

(2)

E.g. in

C^{2}

∣ (2 + 3 i, - 1 + 5 i) ∣ = 2^{2} + 3^{2} + (- 1)^{2} + 5^{2} = 4 + 9 + 1 + 25 = 39

(3)

Normal subgroup Updated 2025-07-16

Ultimate explanation: math.stackexchange.com/questions/776039/intuition-behind-normal-subgroups/3732426#3732426

Only normal subgroups can be used to form quotient groups: their key definition is that they plus their cosets form a group.

Intuition:

One key intuition is that "a normal subgroup is the kernel" of a group homomorphism, and the normal subgroup plus cosets are isomorphic to the image of the isomorphism, which is what the fundamental theorem on homomorphisms says.

Therefore "there aren't that many group homomorphism", and a normal subgroup it is a concrete and natural way to uniquely represent that homomorphism.

The best way to think about the, is to always think first: what is the homomorphism? And then work out everything else from there.