In which @yuanchuan reconstructs the generative images that (used to) accompany OpenAI's blog posts:

https://yuanchuan.dev/daily-sketch-4

Amazing stuff.

My favorite reconstructions are 3 and 15 - which how much can with so little

I trained a neural net.

If you want to, you can too. From scratch. The only thing you need is a 2 hours, a laptop, and a vague recollection of high school calculus. Andrej Karpathy will teach you the rest:
https://www.youtube.com/watch?v=VMj-3S1tku0

#machinelearning

Building up hyperbolic functions from exponents

The Wikipedia article - https://en.wikipedia.org/wiki/Hyperbolic_functions - has great illustrations, this plot was my attempt to make them myself just to grok better

#math

#PyTorch operations involve two dynamic dispatches

From: https://blog.ezyang.com/2019/05/pytorch-internals/

(and what a lovely font)

I've been enjoying thinking about vector spaces so much that I wrote another description of them without using (almost) any mathematical terminology:

An introduction to vector spaces without a single equation

https://mnvr.in/vector-spaces-2

Artwork: Manfred Mohr, Cubic Limit (1977)

#math #machinelearning

1/x is determined by x > 1. Same curve reflected across x = y is 0 < x < 1. Negatives are another reflection.

What we have is a logarithmic number line, centered (instead of 0) at 1 and growing (instead of additively towards ∞ and -∞) multiplicatively towards infinity and the infinitesimal.

#math

Permalink: https://mnvr.in/2026-a#01-03-d

Created a small scrobbler for my website!

It's a small #Python script that'll take the yt URL as a command line arg, fetch its metadata and thumbnail, use Pillow to extract the 3 most dominant colors, and emit a #HTML fragment that I can then paste into my log.

Code - https://github.com/mnvr/youtube-colors

In action - https://mnvr.in/2026-a#01-03-c

#web #music

Vector Spaces for Machine Learning

I wrote about vector space concepts I had to clarify for myself in the context of ML - https://mnvr.in/vector-spaces

Some of you who are interested in #machinelearning might find these useful too!

uv in #webgl means "coordinates on the texture"

xyz are in 3d space
uv are in texture space

usually, they're also normalized, so they are effectively "normalized texture coordinates" in [0, 1]^2

#shaders

#til Adding a semicolon at the end of the last line of a #Jupyter cell supresses output - Useful if we just want the side effects (e.g. drawing a plot)

#Python #machinelearning