2026-04-16 23:17 UTC
A complete transformer neural network implemented entirely in HyperTalk, trained on a Macintosh SE/30.
https://github.com/SeanFDZ/macmind
Replies (3)
-
@triplingual@code4lib.social 2026-04-16 23:19
@kjhealy 😍
-
@tantramar@zeroes.ca 2026-04-16 23:46
@kjhealy “stochastic gradient descent” or GTFO, amiright? (I have no clue what that means.)
-
@MacBalance@mstdn.games 2026-04-18 01:13
@kjhealy SoI watched (part of) a video earlier today of someone doing something similar with an ancient PDP11. (He has it in his garage. I think it’s an eccentric enough old man hobby) Are the “modern” LLMs basically this with a ridiculously larger pool of resources to train on? (Plus, I assume, various shims and rules to try to keep them at least sounding sane.)