getting_started with microgpt

Getting Started with Learning AI, Part 2　Texts and planned by Yoshio Okamoto / Codes and PDF written by Claude Opus5.5 (Anthropic) Octber 23, 2026


Here we dig even deeper into microgpt, which we covered in Part 1

Although microgpt is said to be the simplest possible LLM, the name file used in the example above contains as many as 32,033 names. Following every step of the computation with that many is a daunting task, so here we limit the input file to just four names! As for why we chose these somewhat confusing names, please see the guide below.

In other words, this is an attempt to strip the world's simplest LLM bare by restricting its internal computation to the smallest possible input!

1. What the computation actually looks like

This is an attempt to look at the entire computation process with the input limited to only four names (we will call this the "minimum example").

Names in the minimum example

emma
ava
mia
anna

Moreover, we deliberately chose this set of names because they are easy for the AI to confuse and get wrong (the reasons for choosing these names are also explained in detail in the guide below).

Since the full process is long, it is summarized in the guide below. This is still a trial version, and I would like to revise it based on various feedback.
When I had the AI write the guide, I added the condition that it should be "understandable even by high school students and the general public." Even so, it may be hard to get into at first, but please give it a try.

A Guide to microgpt Trained on Just Four Names (trial version)

https://seagull.stars.ne.jp/2026_LLM/microgpt2/Dissecting_microgpt_en_v0.1.pdf

2. In addition, there is a series of guides on machine learning and deep learning leading up to LLMs, made with the help of Claude so that even high school students can understand them. I myself have not fully understood them yet and am still reading through them bit by bit, but I am publishing them here as trial versions as well.

I would be grateful for any comments. yossi.okamoto [at] gmail.com

<Machine Learning>
Supervised and Unsupervised Learning
https://seagull.stars.ne.jp/2026_LLM/Machine_Learning/EG/Supervised_and_Unsupervised_Learning_HS_en_v0.1.pdf

<Deep Learning>

Deep Learning (1) Basics of Neural Networks
https://seagull.stars.ne.jp/2026_LLM/Deep_Learning/EG/DeepLearning1_Basics_of_Neural_Networks_HS_en_v0.2.pdf

Deep Learning (2) Backpropagation and Training in Practice
https://seagull.stars.ne.jp/2026_LLM/Deep_Learning/EG/DeepLearning2_Backpropagation_and_Training_HS_en_v0.2.pdf

Deep Learning (3) Convolutional Neural Networks
https://seagull.stars.ne.jp/2026_LLM/Deep_Learning/EG/DeepLearning3_Convolutional_Neural_Networks_HS_en_v0.2.pdf

Deep Learning (4) Sequence Data and Transformers
https://seagull.stars.ne.jp/2026_LLM/Deep_Learning/EG/DeepLearning4_Sequence_Data_and_Transformers_HS_en_v0.2.pdf

<LLM>

LLM Basics (1) Tokens and Language Models
https://seagull.stars.ne.jp/2026_LLM/LLM/EG/LLM_Basics1_Tokens_and_Language_Models_HS_en_v0.1.pdf

LLM Basics (2) Pretraining and Scaling Laws
https://seagull.stars.ne.jp/2026_LLM/LLM/EG/LLM_Basics2_Pretraining_and_Scaling_Laws_HS_en_v0.1.pdf

LLM Basics (3) Toward_Chat_Models
https://seagull.stars.ne.jp/2026_LLM/LLM/EG/LLM_Basics3_Toward_Chat_Models_HS_en_v0.1-1.pdf

Please be sure to read the guides above together with videos such as the following.
These Guide sites only in Japanese, please look for English sites.

The YouTube video series by 3Blue1Brown Japan (https://www.youtube.com/@3Blue1BrownJapan)

Yobinori's YouTube video series explaining deep learning (https://www.youtube.com/watch?v=s5_Pk3CjhNA)

3. Other brief explanations of microgpt itself (earlier material; PDFs proposed by Okamoto and created by Claude)

https://seagull.stars.ne.jp/2026_LLM/microgpt/Reading_microgpt_en_v0.1.pdf

How the matrix calculations inside it work (supplement)
https://seagull.stars.ne.jp/2026_LLM/microgpt/Reading_microgpt_tables_en_v0.1.pdf