Pylimitics

Everybody knows a lot more than they’re able to explain. There are some areas where language just isn’t adequate. For example, imagine trying to explain an aroma without referring to other smells. Without “it smells like…”, there’s not a lot you can do.

Even though people have amassed countless pages of written information, and that’s sometimes called the “total of human knowledge,” it isn’t even close. It’s only the part we can express in language. The stuff we can explain.

It’s a fantastic achievement when a writer manages to communicate or invoke or somehow conjure up something that is known but seemed impossible to explain. Everybody still knows the names of historical writers who did this: Shakespeare, Tolstoy, Ginsberg, Dickinson. There have been a lot of them, although the number is tiny when you consider that these are the standouts among everybody who has written.

We still read and study the works of writers like those. Their breakthrough successes extend the capability of language to explain something you already know but couldn’t express. Sometimes the shorter and pithier of their innovations actually enter the language. “Wild goose chase” and “all that glitters is not gold” are both from Shakespeare. “The pen is mightier than the sword” is from Edward Bulwer-Lytton (who is less revered these days).

I would argue that our ability to read and recall and reuse the breakthrough insights of the best writers from history mean that as ordinary people we can express more of what we know than was the case for people from earlier centuries.

What does it mean that we know something? It’s clearly not simply that we remember the words. It has something to do with the complex total of our senses and our minds. When we can express part of that, it’s a bonus, but it’s not the same.

But imagine someone who is phenomenally good at remembering the words and how they’ve been used by others, but has no experiences represented by the words. Maybe they have few or no senses other than what they need to learn and repeat words. They are alive, in a living body, but maybe without sight, taste, smell, touch, or hearing.

We might have conversations with that person, and they would be able to use their great knowledge of words to say things that make sense to us. But there would be nothing behind the words. From that person, the words don’t represent much of anything. Is that hypothetical person really expressing anything? Or just recombining the words that they remember extremely well but can’t connect to anything; they have no experiences. if the words make sense, that sense lies entirely within us. The words, in this case, are not communicating experiences to us.

A large language model is the closest thing I know of to that hypothetical person with no senses, no experiences, but great memory for words and how they’ve been used by people. An LLM is a repository of writing — a vast amount of writing. All that writing was done by people who found ways to express what they knew, sometimes in ways so original we still use them. The LLM can say things that make sense to us. But there is nothing behind the words; they don’t represent much of anything.Iif the words make sense, that sense lies entirely within us.