View on GitHub

AI / ML Dev Log

Reverse chronological dev-log of my journey trying to learn modern AI / ML topics, including collections of links and ideas I’ve come up with along the way.

Learning Resources



My explorations around an AI toolkit utimately fizzled out, and I redirected my attention to Rust and Ray Tracers. The Python-based ML projects were ultimately unsatisfying - Python is a frustrating language to work with, and I was mostly using libraries instead of understanding concepts. Applying ML tooling can be useful, but I prefer to be able to build things myself - even if they won’t match modern libraries, having a foundational and engineering-based understanding of modern ML is within reach. Recently I was inspired by the llama.cpp project, which can run Meta’s LLaMA models on a CPU with ChatGPT-like performance. The entire implementation is about 20k lines of C and C++, which points to how accessible the core algorithms are - that includes support for various SIMD platforms, so the core inference logic is likely less than a few hundred lines of core mathematical code.

To continue my learning here, I’m starting out from the other end - building a simple Tensor library in Rust. Coming from this direction builds off some existing engineering skills, versus trying to learn the ML and mathematics which has felt more abstract and inaccessible. I’m looking forward to diving into lower-level code (e.g compute shaders and SIMD instructions), and hopefully building towards something capable of doing inference of a (yet to be determined) simple ML model.



Depth map of living room

Broken depth map of living room

Profile photo depth map



original fish swimming through water

First is the original image (run through SD with low strength for added anonimity) and bottom is the edited. You can see some vague references to water - the carpet, the TV, and the window - but also you can clearly see a TV and a sofa and coffee table - and most notable of all, no fish in sight.