hexhowells

The human ability to learn from few examples can be explained with evolution (and thus search). We evolved to be fast learners as it was key to our survival. If you touched fire and felt pain, you better learn quickly not to keep touching it. This learning from reward signals (neurotransmitters) in our brain generalises to pretty much all learning tasks

> Like a lot of human actions in space, folding clothes and other motor tasks are hierarchical sequences of smaller tasks that are strung together

I disagree, you can model those tasks as hiearchical sequences of smaller tasks. But the terminal goal of folding clothes is to turn a pile of unfolded clothes into a neat pile of folded clothes.

The reason you would break down the task is because getting between those two states with the only reward signal being "the clothes are now folded" takes a lot of steps, and given the possible actions the robot can take, results in a large search space.

OP here, I whipped this up in like 10 minutes after modelling the problem from a new perspective (I want to be less of a perfectionist with my blogs) so there are definitely grey areas I didn't consider/cover.

I do think LLMs can be good for certain boilerplate code whilst still allowing you to enjoy the problems you care about, and as far as my binary definitions this is more of a grey area.

I guess for me, this has introduced a slippery slope where if the LLM can also code the "fun" stuff, I'll be more inclined to use it, which defeats the whole purpose for me. Perhaps being able to identify which type of project I am working on, it can help me avoid using LLMs to enjoy programming more again!

Cool project! I worked on something similar a while ago to learn about automatic differentiation: https://github.com/hexhowells/onegrad.js

Needs more layers to be really useful though.

I always thought examples using browser extensions would be neat since their built in JS and you only need to download the model once.