LLM From Scratch offers free course to build tiny AI models

8 articles · Updated · XDA Developers · May 8

The six-part workshop covers tokenisation, transformer architecture, training, inference, scaling experiments and a poetry challenge, and is designed to run on a laptop.
It takes users from basic character-level tokenisers to a working small language model trained on poetry, aimed at hobbyists rather than competing with systems like ChatGPT or Gemini.
The course materials and details are available on the project's GitHub page, offering a practical introduction to how large language models are built from scratch.

With 2026 optimizers, can a laptop-built AI poet truly rival the creative output of large commercial models on specialized tasks?

As local AI agents become common, what are the key security risks of running a self-built LLM on your personal laptop?

Is building an LLM from scratch an educational rite of passage or an inefficient detour from fine-tuning superior pre-existing models?