SeanFDZ

SeanFDZ / macmind

Public

Single-layer transformer in HyperTalk for the classic Macintosh

attention backpropagation classic-mac fft hypercard
27
2
100% credibility
Found Apr 16, 2026 at 27 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

An interactive HyperCard stack on vintage Macintosh that lets users train and test a tiny transformer model to learn rearranging number sequences, with visualizations of its inner workings.

How It Works

1
🔍 Discover MacMind

You come across a delightful project that brings a tiny thinking machine to life on a 1980s Macintosh, showing how AI really works.

2
📥 Download the Fun App

Pick up the ready-to-play version of the interactive card stack from the downloads.

3
🖥️ Launch the Cards

Double-click to open the colorful stack in your vintage Mac setup and flip through the simple cards.

4
🧠 Test the Magic Rearranger

On the testing card, create a random row of numbers and press go to watch it rearrange them perfectly every time.

5
📊 Reveal the Secret Pattern

Flip to the picture card to gaze at the beautiful web of connections the thinker discovered all by itself.

6
🎯 Watch It Learn

Head to the training card, tap the buttons, and see progress bars fill as it gets smarter with each lesson.

✨ AI Makes Sense Now!

You walk away amazed that the same idea behind huge modern brains runs on tiny old hardware—pure math magic unlocked.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 27 to 27 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is macmind?

MacMind runs a single-layer transformer for language modeling entirely in HyperTalk on classic Macintosh hardware, training it to predict bit-reversal permutations—the key first step in Fast Fourier Transforms—from random digit sequences. You get an interactive HyperCard stack for real-time training, inference on 8-digit inputs, and attention map visualization showing how it discovers FFT butterfly patterns. Python provides a NumPy reference to validate results match modern single-layer transformer math.

Why is it gaining traction?

It demystifies transformers by making every weight, gradient, and attention score editable in a 1987 scripting environment, with no compiled code or libraries—pure inspectable math on a 68000 CPU. Developers love tweaking hyperparameters via HyperCard's editor and watching backprop grind on emulated vintage Macs, revealing how tiny models bootstrap complex patterns like those in sgformer or single-layer vision transformers. The retro AI stunt hooks retrocomputing and ML crowds seeking tangible intuition over black-box scale.

Who should use this?

AI educators teaching attention and backprop in workshops, retro Mac enthusiasts running System 7 emulators like Basilisk II, or ML engineers grokking single-layer transformers before building larger ones. Ideal for visualizing mcmindfulness in toy tasks or experimenting with HyperTalk as a constraint for mechanistic interpretability.

Verdict

Fun, educational demo with excellent docs and Python validation, but low maturity at 27 stars and 1.0% credibility—great for tinkering on emulators, skip for anything practical. Train it overnight for a humbling reminder that transformers are just math, Macintosh or not.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.