Feeling good enough about this to step back from it for a little bit now.

Now with a table overlay... looks nice, I think. But the table overlay does not exactly scream interpretability.

Lots of cool takes on process in these slides.

From: https://twitter.com/hypotext/status/874063290040700929

We're going full width.

This is, I think, the best version of this direction, but I think I am going to change direction.

It was fun to figure out the math for a circle. But I think it's probably not the right option here.

This seems promising. This is maybe mixing a couple of different aesthetics, need to think through how they mix some more.

Curves didn't look good. Mocked up the rest of the cover to get a sense of proportion. Feeling optimistic. Going to think more about how to do the one color gradient.

This is more at the scale that should look good on a book. Going to try some curves now.

A promising start to the new direction.

Maybe this is a better way to get the complex neural network point across for the cover illustration.

From: http://terencebroad.com/nnvis.html

Connected.

Regions now being filled with circles programmatically.

Region subdivision working.

Every dot connected to every other dot.

Starting work on the cover. The idea is a black box that is revealed or half-revealed. I am going to try making the inside programmatically with d3 and/or react.

Lots of figures in the second chapter of the interpretability report.

Interpretability report figures in progress.

Strangely compelling.

From: https://twitter.com/M_PF/status/869982898153246721

Working on figures for the interpretability report.