AI by Hand | Hacker News Reader

AI by Hand

byhand.ai

367 pointssans_souse31 comments

Open articleView on HN

This is also recommended reading if you want to learn more about building LLMs from scratch: See:

Train your own LLM - https://github.com/angelos-p/llm-from-scratch

HN Discussion - https://news.ycombinator.com/item?id=48017948

Are there any like this but in c or c++?

https://github.com/karpathy/llm.c

However, I wonder about the use of this. Typically, you'd want to use a GPU for inference, and programming those in C is quite a challenge. The educational value of learning LLMs in C/C++ seems a lot lower than in Python. May I ask why you're interested in this specifically?

Also, why ask this here, and not simply use Google or an LLM to find the information? It comes off as a bit lazy, especially given that you provide no extra context.

Username checks out

> Also, why ask this here, and not simply use Google or an LLM to find the information? It comes off as a bit lazy

You gave a good answer: why SO the thread? Conversation is the goal here, right? If you’re feeling put out by the question, why bother answering it and then complaining that they asked it? Just move along.

> Also, why ask this here, and not simply use Google or an LLM to find the information? It comes off as a bit lazy

- why answer?

- Google is a fucking mess, no thanks, quicker to ask a person who knows the answer already

- llms don't always give you a decent answer, and the time I spend back and forth would be greater than just asking a human

Why c? Because I want to hand craft a minimal llm without calling in a bunch of libraries that do the actual LLM part for me... I want to know the actual maths going into it. Plus I don't know python. By the looks of it, you can code an llm ina few lines of correctly chosen python; to me this is very unenlightening and I might as well have just ran ollama

https://www.byhand.ai/p/library

> AI by Hand is the research publication of By Hand Research, founded by Prof. Tom Yeh. By Hand Research studies model interpretability and explainability at the math and algorithm level. Subscribers receive free new articles and join live seminars. Members get access to our full research library.

I'm not sure what this is. Click past the subscribe page to get a bunch of links to descriptions of articles I could read if I was a subscriber?

Click "Later->" to bypass subscribe.

dont see later

Eventually, none of us do.

I clicked "Back" to leave the user-hostile website.

Yes. Convince me why your site is worth my attention, don't try to force it through a dark pattern and then try to justify why I should waste my time figuring it out.

Welcome to Substack

You’re safe now. Good job.

Scroll all the way down and there's a video, it's a course he's making for deep-dive into how Qwen 3.6 works internally

- half the posts are locked

- other half have no visualizations whatsoever

- i have posted far better resources than this in the past collecting everything HN has over a 10 yr period on these subjects https://news.ycombinator.com/item?id=49183198

- EDIT

- every post is locked https://www.byhand.ai/archive

- sorry bruh, not gonna bother with a paywall that big

- amaze me first, i am happy to pay

I found Deep Learning by No Starch Press to be an excellent primer the foundation of AI:

https://nostarch.com/deep-learning-visual-approach

There's also The Little Learner: https://www.thelittlelearner.com/

i created something similar a while back. Inspired by micrograd for showing the connection between math/calculus and code, then building along the way to a full NumPy deep learning library that I pretrained GPT-2 124M model with it. One way to learn is to trace through the PRs merged to the repo in chronological order.

The philosophy is the same: "What I cannot create, I do not understand." https://github.com/workofart/ml-by-hand

wow this is nice work. i am diving into this world and was hoping to this exact thing. i am going to look to see how you did it. ty for sharing.

It started mostly as an exercise to make sure I actually understood forward propagation and backpropagation rather than just using an existing framework.

If anyone is interested: https://github.com/kvladsrc/gym/tree/main/neural_network/nn_...

Similarly, I’ve been working through Raschka’s Build a Large Language Model (From Scratch) https://sebastianraschka.com/llms-from-scratch/ and it’s been well worth it (particularly paired with his explainer videos)

My professor used this "by hand" technique when teaching quantum computation algorithms, so I have a soft spot for it.

Nice! Is that also accessible online?

No :(

I’d say bad UX design. It may or may not be something good behind the door. But why do I enter?

I watched the intro video and all I saw were numbers in an Excel sheet. That doesn't make this fun. I'd rather have AI ELI5 it to me.