About
AI Like I'm Five
My learning log for working through 30papers.com, the reading list Ilya Sutskever reportedly gave John Carmack. Written first for an audience of me: I'm autistic with ADHD, and these notes are built to make sense to my brain. If they make sense to yours too, even better.
What gets published here
This blog is my learning log. I'm working through 30papers.com, the reading list Ilya Sutskever reportedly gave to John Carmack, and writing down what I understand as I go. We've finished paper #1 (CS231n, a Stanford course on convolutional neural networks), paper #2 (AlexNet, the 2012 paper that won the ImageNet contest with a deep network), paper #3 (ResNet, the 2015 paper that made networks more than a hundred layers deep trainable), paper #4 (Identity Mappings, the 2016 follow-up that looks at why ResNet's shortcuts work), paper #5 (Dilated Convolutions, the 2016 paper that labels every pixel of an image instead of the image as a whole), and paper #6 (Andrej Karpathy's 2015 post on recurrent neural networks, which read text one character at a time), and are now on paper #7: Understanding LSTM Networks, Christopher Olah's 2015 explanation of the network design behind paper #6's experiments.
Each post breaks down one concept from the reading, built from first principles. No prerequisites. These are notes, not a textbook.
Who it is written for
Primarily: me. I'm autistic with ADHD, and these posts are structured the way my brain needs them: one concept at a time, everything defined, nothing assumed. To some readers that may come across as overly simple or dumbed-down. That's by design. If the notes help you too, great; that's a bonus, not the goal.
How it is written
I explain the math when it matters and skip it when it doesn't. If something is genuinely hard, I say so. Corrections and updates are made in place, with the date shown at the top of the post.
Who writes it
- Richie Thomas, Engineer (45 posts)
Get in touch
Questions, corrections, and requests are welcome through the contact page.