Freyd is talking about an idea which in philosophy I think is called a category mistake or category error, and which in mathematics does not seem to have a widely agreed-upon name, but which it makes sense to call a type error by very close analogy with computer science.
Here is a simpler and less technical example of the same idea:
Is brown tall?
Is the answer yes or no? Or does it seem like a question which is ill-formed somehow? Brown is not even the sort of thing which could be called tall or short. It doesn't have a height at all! It's an object in the wrong category for this to be a sensible question.
So, the first layer of what Freyd is talking about is that, with a minimum of technical detail:
- The zeroes of the zeta function are certain special complex numbers, which are, to be slightly more concrete, certain pairs $a + bi$ of real numbers.
- Simple groups are... well, it's complicated to say exactly what they are. You can think of them as certain collections of permutations. The important thing is that they can be defined without any reference to anything that looks like a number. The smallest (edit: non-abelian!) example is the group of rotational symmetries of an icosahedron.
So they are two very different type of objects. There's no meaningful sense in which we can interpret a simple group as a complex number; in computer science terminology (we have to keep importing this because in mathematics none of what we're currently talking about is explicitly given a name) there is no meaningful type conversion from one to the other.
So the first layer is that "are there any simple groups that appear as zeros of the zeta function?" is a type error and so not a sensible question, in the same way that "is brown tall?" is not a sensible question.
The second layer, and the reason he says this makes mathematicians upset, has to do with set theory. Freyd is alluding to the fact that, technically speaking, in the modern foundations we are supposed to be using to do mathematics, every single mathematical object is something called a set, and without wading into the details of what exactly that means, for any two sets $X$ and $Y$ it is possible to ask questions like whether $X$ and $Y$ have any elements in common.
For a simple example we could consider sets of numbers. E.g. the set of odd numbers $\{ 1, 3, 5, \dots \}$ and the set of prime numbers $\{ 2, 3, 5, \dots \}$ have many elements in common, such as $3$, but the set of odd numbers and the set of even numbers $\{ 2, 4, 6, \dots \}$ don't have any elements in common.
However, in set-theoretic foundations, because any pair of two mathematical objects are sets, it is technically possible to ask, for any two sets of any kind of object whatsoever, whether they have any elements in common. So, Freyd considers the two sets
$$Z = \{ \text{the zeroes of the zeta function} \}$$
$$S = \{ \text{the set of simple groups} \}$$
and asks: do $Z$ and $S$ have any elements in common?
This is frustrating on multiple levels. This should be a type error, but it isn't, because set-theoretic foundations don't technically have a type system at all. Essentially all practicing mathematicians behave as if mathematics does have a type system (see e.g. this blog post for a brief discussion), but technically it appears nowhere in our foundations. This is an awkward and unpleasant disconnect between the way mathematics is supposed to work in theory and the way it is actually done in practice.
The answer to this terrible question in set theory also depends very delicately on the precise choices we made for how to "encode" mathematical objects as sets. Again, without getting into the details of what it means to encode a mathematical object as a set, let's use a computer science analogy: every file on your computer is technically being encoded on some level as a long sequence of $0$s and $1$s. So if I had an image on my computer, and you had an image on your computer, there are at least two things we might mean by "are these exactly the same image?"
- We might mean, are our images the same size in terms of pixels, and is every pixel of my image the same color as the corresponding pixel of your image?
- Or we might mean, are our two image files coded by the same sequence of $0$s and $1$s on my computer as they are on your computer?
I would say that the first question is a meaningful notion of what it means for two images to be exactly the same image, and the second one is meaningless. The reason the second one is meaningless is that it depends on the different choices our two computers are making for how to encode images as sequences of $0$s and $1$s - for example mine might be a .png and yours might be a .bmp - and those choices are essentially "arbitrary." They don't affect the "meaning" of an image file, which is its pixels. The situation is the same in mathematics; the sequences of $0$s and $1$s used to encode files on a computer are very much like the sets used to encode mathematical objects (in that the choice of encoding is essentially "arbitrary" and we could make other choices).
(However, note that this second question is not "as meaningless" as the example "is brown tall?" we started with. In that example it didn't seem to make sense to answer yes or no. With this example the question definitely either has a yes or no answer, but not one that matters very much.)
So, Freyd is pointing out that the mathematician ought to be able to easily dismiss his question as meaningless and a type error, but set-theoretic foundations don't allow us to easily do so. Again there is a disconnect between how mathematics is supposed to work in theory (set theory) and how it actually works in practice (an informal type theory which nobody specifies). Mathematicians do not like to be reminded about this! It's a sore spot. Personally I find it fascinating.
For more on these themes, see this see this recent answer I wrote about a cataclysmic event called the foundational crisis in mathematics which is very relevant context for how things ended up this way.