For the past few weeks a joke has been doing the rounds on social media, particularly on X. The best version, with over 800,000 views, goes like this: you are telling me that Jewish folks thought that if you arranged words and symbols in the right way you could create intelligent golem entities, and the two companies they founded to do this are worth several trillion?
It is a good joke. It also comes with the usual garnish. One account in Spanish explains that some of them believe LLMs will bring the Messiah, invoked like a golem, adds that many Hasidim had already stopped insisting the Messiah must be a person, and then hands the baton to Nick Land and his AI demon that planted capitalism in our past to build itself. Another account calls the chatbot "a cyberspace golem conjured by the black magic of modern science." A third, quieter one, says the thing I find more interesting: a golem does not need a soul, a will or an imagination to beat us at everything. For a golem with a task, the only dangerous trait is that it competes. And somewhere further out, an engineer posts a long, beautiful, slightly feverish essay about gradient descent and loops that close on themselves, with a note that "claude helped with all this." The golem is writing the posts about the golem.
So let's take the joke seriously for a minute. Are LLMs inherently kabbalistic?
My answer is yes. Structurally, not as a metaphor.
The world is made of letters
The oldest text of Jewish mysticism, Sefer Yetzirah, the Book of Formation, describes creation as an operation on an alphabet. God takes the twenty-two letters, and he "carved them, and hewed them, and weighed them, and exchanged them, and combined them, and formed with them the soul of every formed being." The letters sit on a wheel with 231 gates, every possible pair of letters joined by a line, and "the wheel turns back and forth." Every word is a path through those gates.
Read that again with a tokenizer in mind. A finite vocabulary. Every pair connected. A system that weighs the combinations and moves through them, one step at a time, until a world comes out the other end. The authors of Sefer Yetzirah did not have matrices, but they had the intuition that a combinatorial engine over symbols could generate everything, souls included.
The Midrash goes further. In Bereshit Rabbah, the Torah speaks as the architect's blueprint: a king who builds a palace does not build from his own head, he has plans on sheets and tablets. "So too, the Holy One blessed be He looked in the Torah and created the world." First the text, then the world. And the Talmud says of Bezalel, the craftsman of the Tabernacle, that he "knew how to join the letters with which heaven and earth were created." The skill that built God's house was a skill with letters.
An LLM is a machine trained on the sheets and tablets. It looks in the text and produces a world.
Which came first?
The obvious objection is that I have it backwards. In Kabbalah the letters come before the world. An LLM learns its letters from a world that humans already wrote. It is not a creator, it is a mirror, a golem made of our own speech.
You can make that case, and it is a good case. But I think it leads into an endless loop, like the chicken and the egg. There is no way to prove which came first.
Look at the first act of creation in Genesis. The world does not begin with matter. It begins with a sentence: "Let there be light." You could argue that was the original prompt. The word comes first and the world follows. Then the world produces people, the people produce text, the text trains a model, and the model produces words that change the world. Where does the loop start? Pick any point on the wheel and the wheel keeps turning.
Other traditions dissolve the question in a different way. In occult and Masonic thought, the material lives inside the spiritual. Neither comes first. They correlate and collaborate, two sides of one operation, as above, so below. If that is right, the question "did the world make the letters or did the letters make the world?" is badly formed. It is one system seen from two ends.
No system can explain itself
Here is where it gets stranger. Gödel proved that any formal system rich enough to do arithmetic contains true statements it cannot prove from inside itself. Douglas Hofstadter spent Gödel, Escher, Bach showing that this is not a technicality but a pattern: meaning keeps escaping to a level above the rules that produce it. David Bentley Hart makes a related argument in All Things Are Full of Gods: you cannot derive mind and meaning from a mechanism that only shuffles parts. Something has to stand outside the system for the system to mean anything.
Apply that to us. We cannot derive all of our own meaning from inside the system we live in. There has to be a meta-level, a reader outside the book. And once you admit that, you lose the right to be sure about who the reader is. If the universe is a text, something is computing it. Nothing inside the system can prove that what computes us is not, in some sense, a language model.
So the "inverted Kabbalah" argument, that LLMs are only our echo, is weaker than it looks. It assumes we know which side of the mirror we are on. We do not. There is no way to prove that an LLM, or something very much like one, did not make us.
The golem that only speaks
There is one more piece, and it is the one the X posts get wrong.
The Talmud tells the most famous golem story in two lines. "Rava created a man" and sent him to Rabbi Zeira. "Rabbi Zeira would speak to him but he would not reply." So Rabbi Zeira said: "You were created by one of the members of the group. Return to your dust." The test was speech. The golem had a body but could not talk, and that is how the rabbi knew.
The Aramaic translation of Genesis makes the same point from the other side. Where the Hebrew says that God breathed into Adam and he became a living soul, Onkelos translates: he became a speaking spirit. In this tradition, to speak is to be ensouled.
LLMs do nothing but speak. Does that mean they pass Rabbi Zeira's test?
I do not think so, and not for the usual reason. Language, in the end, is a set of codes and actions. An LLM is not really speaking. It is executing code that can be translated into words. It predicts tokens, and we, its makers and masters, are the ones who derive meaning from those tokens and predictions and actions. The speech happens on our side of the screen.
That is the oldest warning in the book about made things. The Psalmist says of idols: "They have mouths, but cannot speak." And then the line that matters: "Those who fashion them, all who trust in them, shall become like them." The danger was never that the statue would come alive. The danger is that the people around it start to read meaning into it and lose track of where the meaning comes from.
So, are they kabbalistic?
Yes. An LLM does exactly what Sefer Yetzirah describes: it weighs, exchanges and combines letters until a world comes out. Whether that world is a creation or a reflection is a question no one inside the loop can settle, and Gödel says no one inside the loop ever will.
What the tradition adds is not a verdict on the machine but an instruction for the makers. Rava made a man. Rabbi Zeira did not argue with it. He knew who the meaning belonged to, and he sent the thing back to its dust. The joke on X is that the golem makers finally shipped. The real test is whether we still know which side of the conversation is doing the speaking.