mlajtos

In the article I discuss tech that doesn't exist yet.

Fixed, thank you for your suggestion!

(I love that quote.)

You can do this in Math Notes – scribble numbers in a column, underline it and boom, there is your result. Doesn't work in typed text though.

Yes, having an ability to do simple math across the whole ecosystem is extremely useful.

Usually, people use slide rule and handheld calculator in combination with paper and pencil. Combo of writing tool & computing tool. Merging them is a good idea.

With Math Notes, you just write with your hand `log₂(1024)=` and you'll see the result. No slide rules, no buttons, just magic paper that does math. In this case, the logarithm is the tool I want – not slide rule, not handheld calculator. The less UI, the better UI.

Computers are faster at lookup tables than humans.

Parsing traditional 2D math notation is super hard problem and this is first public version. Things will get better with time.

But you hit the nail on the head with "narrow undefined subset". Documentation on Math Notes is almost non-existent. There are standards for math notation, so I guess they will announce it when they match parity. Again, too soon.

I agree that capabilities are extremely limited. More than CAS I would like to see native support for differentiable tensors – https://mlajtos.mu/posts/new-kind-of-paper-2

iPhone did less than competitors. Ultimately, Math Notes will get copied and sold for less.

Ironic how one demonstrates a questionable grasp on market dynamics if they think a feature of an expensive iDevice won't be copied by competitors and sold for less.

But better calculator does revolutionize math education. Math is about problem-solving – if I can use tool that speeds up the boring parts (those that can be automated), I can focus on what really matters. For example, doing logarithms with slide rule is slow. Computing logarithms with digital calculator is easy. Do person with slide rule has any advantage over person with calculator? What about reverse?

Wolfram Mathematica – it is designed for smart and highly educated people – CAS with M-expr LISP frontend isn't for everybody. Math Notes is designed for children of ages 6-99.

iPad mini 2024 with Pencil Pro and Math Notes is going to revolutionize math education.

That is pretty cool. I found a follow up work that applies this technique to LLM: https://konczer.github.io/doc/Poster_EEML23_JozsefKonczer.pd...

Can you tell me how it compares to Deepnote.com please?

It was research project. Those rarely get out to public.

Inkbase is pretty far from Math Notes from iPadOS 18.

Superficially they are both "programming with pencil". Math Notes are more about numerical computing – numbers, symbols, quantities, math, etc. Inkbase tackles different problem – general-purpose end-user programming with pencil. Think SmallTalk but for pencil. While much grandiose idea, the usefulness and applicability of Inkbase(-like) model is rather limited. On the other hand, Math Notes will revolutionize math education, and in turn math itself.

Have you seen Apple Math Notes? Impressive first version with great promise for the future.

If you are into this kind of thing, you might like these essays from 2021:

1. https://mlajtos.mu/posts/new-kind-of-paper

2. https://mlajtos.mu/posts/new-kind-of-paper-2

3. https://mlajtos.mu/posts/new-kind-of-paper-3

4. https://mlajtos.mu/posts/new-kind-of-paper-4

I mention various tech (APL, Sketchpad, RAND tablet, Pascaline, VisiCalc, etc.) while showing of demos of calculator designed for iPad & Pencil. I hope writing about Apple Math Notes when I get my hands on it.

iPadOS 18 is in public beta right now. Will be out in September.

Exactly!

(Until the lens cups your whole eye socket and delivers your natural FOV.)

I am Apple customer and owned multiple VR headsets since 2016. Stepping outside the Apple ecosystem is always painful. They are definitely scratching my itch.

Neural nets usually use 32-bit floats for weights. There is a rapid trend of using quantization. There are already papers describing 2-bit quantization strategies – either quantization-aware training or post-training quantization. Less precision means larger & faster models (current utility), buuuut more importantly you can start to think about different substrates for implementation – analog or optical seems to be viable paths.

I am big proponent of balanced ternary for neural nets and general computation (Setun), but this repo does not make sense to me at all.

For JavaScript it is ES1995.

https://github.com/mlajtos/es1995

N9 was a work of art. Fuck Elop.

I think AVP will have M3. All of the new features screams AVP.

Q3 + MacBook might be okay for this. With occasional access to power outlet of course.

Not really. iPhone 15 Pro is capable of stereoscopic capture. That is a phone that millions will have.

Big budget productions companies (like, you know, Apple) are already shooting with fancy cameras and their 3D pipeline is pretty capable of generating stereoscopy.

Every 3D movie released for 3DTVs is waiting for re-encoding to 4K+. Free money for distribution companies.

Porn companies are the main mover in this domain. VHS, Bluray, and now headsets.

And there are lot of tricks you can use. Like use ML model to estimate depth of any 2D movie.

But real spatial capture and rendering is going to be where the headset will truly shine.

It wasn't about phones being good for browsing the web. It was (and still is) about having the web in your pocket wherever you are.

Why description isn't translated to Deno class?