Oooooooooh. Now I get it. I even read
Without catching on.
Now that everybody is (probably) up to speed on this one, I’ll say a bit more. The lower right illustration kind of breaks the premise. You wouldn’t see that angle notated if the book illustration was really supposed to be a rectangular prism. All the corners would be implicitly assumed to be right angles, or they’d have that little standard right-angle corner symbol.
You wouldn’t have the a and b notations (or the r and d ones) either - but of course the shading and other tricks lead the viewer to ignore those too on first glance.
Both of those could be 3D figures with elliptical cross sections though.
Couldn’t the last one be a parallelepiped (in which case the angle would be non-right?)
If we’re not meant to assume that all the angles are right, then there would be at least two angles that would need to be specified, not just one.
You only have to assume that the sides are parallel, which given that it defines the length for only one side, seems like a reasonable assumption.
I like esoteric humor as much as the next guy, but I gotta think the target audience for this joke is really narrow. (I’m not in it.)
I am.
Having had to thrash out Python environments, I would be perfectly willing to sacrifice as many neural nets as needed to get it fixed.
All of the faces can be presumed to be parallelograms. And we can probably also assume that the face closest to us is a rectangle. But that still leaves two pairs of faces, each of which could be at any angle.
Not if opposing sides are also the same length. This is an assumption, but it seems a reasonable assumption given that the length of only one side is defined.
I already posited that all sides are parallelograms, which implies opposing sides being the same length. It’s not enough.
Think of it this way: That diagonal dotted line could be pointing in any direction. Its endpoint is a point in 3-dimensional space, so we need three numbers to describe it. The length of that segment, d, is one of those numbers. The given angle theta is another number. What’s the third point to specify the endpoint?
Oh, and Munroe needs a genetic algorithm. At constant intervals, every surviving AI makes a copy of itself, with slight variations. The ones that are best at attempting to fix his Python environment will last longest before self-deleting, and hence produce more offspring.
Maybe it would be more instructive to train an adversarial network to generate worse Python environments than he already has.
Surely the ones most tolerant of the chaos of his Python environment will survive longest (which is how Randall got into this problem (by being tolerant of the chaos enough at the beginning to let the problem get worse)) 
And they’re assuming a spherical beachball…
That BLM banner at the top of every xkcd comic page so badly triggered an online friend that I decided I didn’t want to be friends with them anymore. I started a thread about it over in IMHO: Is out-of-hand rejection of BLM conclusive evidence of unapologetic racism?
Piña Collider… really, Randall?
A shaggy dog in 3 panels.