After more than 25 years, Cyc now contains 5 million assertions. Lenat has said that 100 million would be required before Cyc would be able to reason like a human does. No significant applications of its knowledge base currently exist

There are; like most GOFAI projects, they tend not to be sexy, but their publications contain some, and they have a few marketing speak writeups of industry users.

In 1969, Rosenblatt was met with a scathing attack by symbolic artificial intelligence advocate Marvin Minsky.

Minsky proved that perceptrons couldn't learn XOR and some other functions, but Perceptrons wasn't a polemic, it just ended up being received as one.

I don't think connectionist and symbolic AI are different in kind; try to classify statistical relational learning as one or the other if you disagree. Somewhere in Paradigms of Artificial Intelligence Programming (I think in the chapter on Eliza) there's a tangent about how AI applications that look magical can be disappointing when you learn the magic trick and there isn't much magic to it. I think it's harder to see the "trick" in connectionist algorithms, because you have a giant matrix and out of your giant matrix comes more or less reasonable responses, and you can't point to how it happened as directly as you can with a proof tree.

And, of course, Douglas Hofstadler argues for a similar conclusion from a symbolic AI perspective.