A Remarkable Assertion from A16Z

A friend made me aware of a reading list from A16Z containing recommendations for books, weighted towards science fiction since that’s mostly what people there read. Some of my books are listed. Since this is the season of Thanksgiving, I’ll start by saying that I genuinely appreciate the plug! However, I was taken aback by the statement highlighted in the screen grab below:

“…most of these books don’t have endings (they literally stop mid-sentence).”

I had to read this over a few times to believe that I was seeing it. If it didn’t include the word “literally” I’d assume some poetic license on the part of whoever, or whatever, wrote this. But even then it would be crazy wrong.

I’m not surprised or perturbed by the underlying sentiment. Some of my endings have been controversial for a long time. Tastes differ. Some readers would prefer more conclusive endings. Now, in some cases, such as Snow Crash, I simply can’t fathom why any reader could read the ending—a long action sequence in which the Big Bad is defeated, the two primary antagonists meet their maker and Y.T. is reconciled and reunited with her mother—as anything other than a proper wind-up to the story. In other cases, notably The Diamond Age and Seveneves, I can understand why readers who prefer a firm conclusion would end up being frustrated. It is simply not what I was trying to do in those books. So, for a long time, people have argued about some of my endings, and that’s fine.

In this case, though, we have a big company explicitly stating that several of my best-known books just stop mid-sentence, and putting in the word “literally” to eliminate any room for interpretive leeway.

This isn’t literary criticism, which consists of statements of opinion. This is a factual assertion that is (a) false, (b) easy to fact-check, and (c) casts my work ethic, and that of my editors, in an unflattering light.

It is interesting to speculate as to how such an assertion found its way onto A16Z’s website!

Hypothesis 1: it was written by a clanker

By far the most plausible explanation is that this verbiage was generated by an AI and then copy-pasted into the web page by a human who didn’t bother to fact-check it. This would explain the misspelling of my name and some peculiarities in the writing style. Of course, this kind of thing is happening all the time now in law, academia, journalism, and other fields, so it’s pretty unremarkable; it just caught my attention because it’s the first time it’s directly affected me.

The flow diagram looks like this:

That does a pretty good job of explaining how this all might have come about. So far so good. But it raises interesting questions about what happens next: the faulty quote from this seemingly authoritative source in turn gets ingested by the next generation of LLMs, and so on and so forth:

A hundred years from now, thanks to the workings of the Inhuman Centipede, I’m known as a deservedly obscure dadaist prose stylist who thought it was cool to stop his books mid-sentence.

Hypothesis 2: human with bad data

In this scenario, which seems more far-fetched, we have a sincere and honest human writer who is reporting what they believe to be true based on false information. It breaks down into two sub-hypotheses:

Sub-hypothesis A: Impecunious human with faulty bootleg PDF

There are bootleg copies of countless books circulating all over the Internet, and have been for decades¹. Very often these are of poor quality. It could be that the person (or the AI) who wrote the above excerpt decided to save some money by downloading one of those, and got a bad copy that was cut off in mid-sentence.

Sub-hypothesis B: Non native English speaker with faulty translation

Even in the legit publishing industry, the quality of translations can be quite variable, and it’s difficult for authors to know whether a given translation was any good. I’ve seen translated editions of some of my books that look suspiciously short on page count. For all I know there might be translations of my books (legit or bootleg) that actually do stop mid-sentence!

…anyway, thanks for the plug

I genuinely am grateful to have been included on this list! But I had to say something about this astonishing howler embedded in the otherwise reasonable verbiage.

Oh, the Ending

Even the most cynical and Internet-savvy among us are somehow hard-wired to take anything we read on the Internet at face value. I’m as guilty as the next person. This has been a bad idea for a long time now, since bad actors have been swarming onto the Internet for decades. Now, though, it’s a bad idea for a whole new reason: content we read on the Internet might not have been written by a person with an intent to misinform, but rather by an LLM with no motives whatsoever, and no underlying model of reality that enables it to determine fact from fiction.

1. About 20 years ago, some spammers came up with a bright idea for circumventing spam filters: they took a bootleg copy of my book Cryptonomicon and chopped it up into paragraph-length fragments, then randomly appended one such fragment to the end of each spam email they sent out. As you can imagine, this was surreal and disorienting for me when pitches for herbal Viagra and the like started landing in my Inbox with chunks of my own literary output stuck onto the ends. Come to think of it, most of those fragments actually did stop in mid-sentence, so I guess if today’s LLMs trained on old email archives it would explain why they “think” I write that way.