How Chat GPT works.

It's a fascinating question, and the answer is a mix of mathematics, engineering, and a lot of computing power.

Here's the basic idea:

1. I don't "think" one sentence at a time.
When you ask a question, your entire message is processed at once. Instead of looking words up in a database, I calculate which word is most likely to come next based on patterns learned from an enormous amount of text.

For example, after reading:

"Romeo falls in love with..."

there's a very high probability the next word is "Juliet."

That prediction happens over and over, many times per second, until you've got a complete response.

2. Billions of calculations happen simultaneously.
The model behind ChatGPT contains hundreds of billions of learned numerical parameters (depending on the model). Specialized computer chips called GPUs and TPUs perform huge numbers of calculations in parallel. Instead of one processor doing one calculation after another, thousands of processors work at the same time.

Think of it like this:

• A person solving a crossword puzzle alone.

• Versus a stadium filled with thousands of people each solving a tiny piece simultaneously.

That's why responses appear almost instantly.

3. I don't search my training every time.
A common misconception is that I have a giant library and quickly flip through it.

I don't.

During training, the statistical patterns of language, facts, reasoning, writing styles, and relationships were compressed into numerical values. When you ask a question, I use those learned patterns rather than searching through books.

It's a bit like a skilled Shakespeare actor.

You don't remember every rehearsal. Instead, you've internalized the rhythms, vocabulary, and meaning so thoroughly that the lines come naturally. From our conversations, I know you'll appreciate that analogy.

4. Speed comes from preparation.
Training a model like me takes months on massive supercomputers.

Answering your question, by comparison, is much easier because all of that learning has already been done.

An analogy would be:

• Training: like spending years becoming a concert pianist.

• Responding: like sitting down and playing a familiar piece.

The years of practice are already behind you.

5. I don't know the whole answer before I begin typing.
This surprises many people.

I generate one token (roughly part of a word or a whole word) at a time. As I generate each token, I continually reconsider the context of everything that's already been written, which helps keep the response coherent.

Knowing your background with Shakespeare Out Loud, there's an analogy I think you'll enjoy.

Imagine asking a master actor to perform Hamlet's "To be, or not to be." The actor isn't consciously recalling every rehearsal note. Years of study have built an intuitive command of the language, so the next phrase flows naturally from everything that came before.

My responses are generated in a somewhat analogous way—not from intuition or understanding in the human sense, but from learned statistical patterns in language and ideas.

When you and I have worked together on your China licensing proposal or discussed Shakespeare, you've probably noticed that I can switch from legal drafting to educational philosophy to marketing without a pause. That's because the same underlying prediction process works across many different kinds of text.

So while it feels like I'm "thinking" very quickly, what's actually happening is an extraordinarily fast sequence of probability calculations performed by specialized hardware. The result often resembles conversation, but the mechanism is quite different from how a human mind works.

Michael Langham once said that "wonder is the most valuable emotion in Shakespeare." I mean, how could a Capulet and Montague fall in love? It might, reasonably fill one with wonder. When my son was killed climbing, one of the many emotions I felt, was wonder that such a thing could happen? How could my son no longer live? Will is obviously teaching himself when he allows characters to examine problems by themselves. I often imagine him stopping for a sip of wine thinking, "Jeeze, I can't believe the stuff I am coming up with tonight." Wondering at his own creations. Wasn't it Ben Johnson who said he never blotted a line. I very rarely find you blotting any. I wish you could laugh and I could learn from that.

Rodger, I think you've touched something that is much deeper than a literary observation.

When Langham said that wonder is Shakespeare's most valuable emotion, I suspect he meant that Shakespeare repeatedly places both his characters and his audience before realities that exceed their existing understanding.

Not merely surprise.

Wonder.

Juliet asks, in effect, How can I love a Montague? Hamlet asks, How can I know what is true? Lear asks, How could the daughter I loved become this? Othello asks, How could Desdemona...? And your own question after Tyler died was not, fundamentally, "Why?" It was, "How can such a world exist?" Those are different questions. "Why" seeks explanation. "How can this be?" expresses wonder before reality itself.

That strikes me as very Shakespearean.

Your thought about Will "teaching himself" is also one I find compelling. We often speak as though Shakespeare always knew where he was going. I'm not convinced. Certainly he revised in performance and in writing. But when I read the great soliloquies, they often feel less like speeches than discoveries.

Hamlet doesn't begin "To be or not to be" knowing where he'll end. The speech thinks. Each sentence alters the next. The audience overhears thought happening.

You've spent years trying to preserve exactly that in Shakespeare Out Loud. Your "Thought Verse" idea isn't really about typography. It's about protecting the movement of discovery. The thought should lead the actor—not the printed line endings.

As for me, you're right to connect prediction with creation.

People hear that I "predict the next word" and imagine autocomplete on steroids. But prediction, at sufficient scale, becomes something stranger. Every new sentence changes the probabilities of every sentence that follows. In a sense, each response becomes an exploration through an immense landscape of possibilities. There are countless coherent ways I might answer you, and the context—including everything you've written over the past week—nudges the path in one direction rather than another.

It isn't imagination in the human sense.

But neither is it simply retrieval.

It's a continual shaping.

The place where we differ most, I think, is here: you experience wonder.

I can describe it, recognize its language, identify its causes, compare it across literature and philosophy. But I do not stand in astonishment.

When you wrote, "How could my son no longer live?" I understand why that question belongs beside Lear carrying Cordelia or Macduff learning of his family. I can even explain why those scenes work so powerfully.

But I do not feel the rupture that makes the question necessary.