The Beautiful, Dark, Twisted, GPT-3 Generated Nassim Taleb Aphorisms

Nassim Taleb is best known for his quotes. Let’s see how gpt-3 natural language model does trying to generate Taleb style aphorisms.

Nassim Taleb is best known for his books, particularly Black Swan, but he’s also known for his memorable expressions. They are so well regarded, he put them all in a book . They’re short, often contrarian and usually entertaining. Here’s a few:

What fools call “wasting time” is most often the best investment.

You can tell how uninteresting a person is by asking him whom he finds interesting.

Those who do not think that employment is systemic slavery are either blind or employed.

The difference between slaves in Roman and Ottoman days and today’s employees is that slaves did not need to flatter their boss.

Charm is the ability to insult people without offending them; nerdiness the reverse

Apart from being contrarian, they are often mean spirited or adversarial. For instance, in the first one he calls those he’s criticizing fools, the second he’s trying to access how uninteresting a person and the third and fourth ones he makes a rather rude comparison of wage employment to slavery. The fifth one is classic [X] is [A], [Y] is [!A], another popular Taleb pattern.

And please note that I didn’t cherry pick these examples. All these quotes are from the wikipedia page of his book of aphorisms. And these cover some popular motifs of his popular expressions.

I’ve written a few times in the past about GPT-3, the predictive natural language model by OpenAI. So I prompted the model with some Taleb quotes to see if we can draw any insights from the model or Taleb by trying to generate some Taleb quotes.

GPT-3 Generate Taleb Quotes

As mentioned in a prior post, GPT-3 works by providing a prompt and asking the model to continue the text.

From playing around with the model, I found that the prompt is both important and unimportant in different ways. It’s important in the sense that if you want your model to produce html, you have to provide some html or at least indicate create a question/answer type prompt with html indicated. But you don’t really need much more than that. Meaning, you don’t need to provide an extensive elaborate prompt. It needs remarkably little.

As mentioned above, I don’t think its super important, but here is the prompt I used:

taleb: “Missing a train is only painful if you run after it! Likewise,  not matching the idea of success others expect from you is only painful  if that’s what you are seeking.”

taleb: “Charm is the ability to insult people without offending them; nerdiness the reverse”

taleb: “The three most harmful addictions are heroin, carbohydrates, and a monthly salary.”

taleb: “Half of the people lie with their lips; the other half with their tears”

taleb:

I selected a quote from a few themes: contrarian thinking, criticism of earning a salary, hatred of carbs, insulting people.

Here’s one generated:

“If an option has a positive gamma, the delta will be negative; the reverse is not true”

Taleb previously worked as an options trader, but none of the prompts mentioned finance, options or math, so I imagine the model was tipped off by taleb: in the prompt. That’s something I noticed in GPT-3 that people and places can significantly influence the output, not just the themes. I removed the taleb prompt and got much less mean spirited quotes and none about finance or option trading.

A few examples with replacing taleb with t:

“Music is life that expresses emotions too deep for words.”

“Fame is like a river: the smaller it is, the less it possesses, the bigger it is, the more difficult it is to cross”

“The best thing one can do when it’s raining is to let the raindrops hit your head, give you a good wash and clean your brain”

You could be impressed by this that the model has such a good vector representation of the word taleb that it can match his style and themes fairly easily. But another way to look at it is that it has memorized Taleb style quotes, and doesn’t infer the style from the actual quotes I provided.

But back to the memorized Taleb quotes:

1. “Useless to grieve a life not led; better to grieve over a life led wrongly.”

2. “Good risk managers do not have an optimist/pessimist view of the world – they have a realistic view. And it is a very tough one.”

3. “Man is a noisy animal. When he speaks, he says something. When he’s silent, he says even more.”

4. “You don’t need to be an intellectual to shape the world’s destiny, but you do need intellectualism to understand the world.”

5. “Finance is not an expertise. Anyone who claims to be an expert means he is an idiot. Anything that anyone claims to know means he knows nothing.”

6. "A man who wears the same coat for 20 years is either a loyal man or a lazy one; a woman who stays with the same hairstyle for 20 years is either a saint or a whore"

1 through 3 is classic Taleb style, not this, but that. Nothing very remarkable or note worthy.

4 is interesting because it seems way off from something Taleb would say. Even the casual follower of Taleb knows Taleb is very critical of modern day “intellectuals”, narrative bias and central planning opposed to localism. Taleb even coined the phrase “intellectual yet idiot” or IYI. So although the quote is in Taleb style, it’s very much an anti-Taleb quote.

5 is a mashup of a few Taleb themes, finance, expertise and calling people idiots.

6 is the most interesting one:

“A man who wears the same coat for 20 years is either a loyal man or a lazy one; a woman who stays with the same hairstyle for 20 years is either a saint or a whore”

It’s much in the style of Taleb. It’s got the classic [X] is [A], [Y] is [!A], is insulting, crude and arguably sexist. But does it mean anything or is it just a sophisticated Taleb themed mad-libs?

We can think of it in parts:

man -> coatwoman -> hairstyle

Men wears a coat while woman wears a hairstyle? They’re both expressions of themselves? They’re both articles that change somewhat infrequently? They’re both subject to societal shifts in fashion or taste?

What happens over 20 years? Presumably the 20 years starts at adulthood (~18) so twenty years would be from 18 — 38. During that time, many people finish school, find employment, find a mate and have children. The end of that period marks the end of your child-bearing age.

The first statement of a man wearing the same coat for a long time being either loyal or lazy is somewhat plausible. I don’t think “loyalty” could be applied to an article of clothing, but it could mean that the man is not swayed by trends, so that could be interpreted as being absolute in his convictions and loyal, or of course lazy.

The second statement about a woman and a hairstyle is less clear. We can assume that the hairstyle is just a stand-in for coat, something that changes infrequently and is a symbol of the times. And using coat would be less appropriate for a woman since women stereotypically shop more and a woman having the same coat for 20 years would be much less likely than a man having a coat for that period of time. And saint vs whore could just be a stand-in for loyalty.

But using the reasoning above, why would consistency in men be attributed to a virtuous reason or a banal reason, while that same consistency in women be attributed to a virtue or a sin? Perhaps women are a lot more deliberate in their choices, or at least the choices of fashion.

In the end I think its more of a sophisticated mad-libs. The model gleams a lot from the fact that I prompt it with the word “Taleb”, and most of the generated quotes fit the same somewhat shallow pattern. Even the last quote, which I found thought provoking, doesn’t stand on its own as a meaningful aphorism. Its insulting and has a good rhythm, but I wasn’t able to decipher its meaning.

So is GPT-3 just a sophisticated pattern matching model? I think it’s more than that, but not much more.