Dylan's Like a Rolling Stone

Dylan’s Like a Rolling Stone

There is a romantic image we have of how art flows out of the poet's mind. Bob Dylan, legal pad on his knee, cigarette in the ashtray, the sounds of Twenty-Third Street coming through the window. The song arriving in a single volcanic session. Six minutes of fury and precision, resulting in “Like a Rolling Stone.”

It didn't happen that way.

Here’s what we do know. Dylan came back from his 1965 England tour exhausted. He later said he was ready to quit music. What came out of him was not a song. It was twenty pages of what he called "vomit," a long, rageful, formless rant. He wrote it in Woodstock. He condensed it over days, cutting and reshaping until something emerged from the wreckage that had a structure he recognized. Then he brought it to Columbia's Studio A, where the band found the rhythm and where Al Kooper was not even supposed to be playing organ that day.

The rant became the song, but not through solitary genius. It took exhaustion, then overflow, then editing, then collaboration, then accident.

The romantic image is cleaner. It confirms what we want to believe about how great work gets made: one extraordinary mind, alone with its vision, producing something that changes the culture. The myth of the author as origin. The source of the river.

All due respect to Dylan – what he produced was extraordinary – but the myth of the solitary genius is an illusion of ego. This masterpiece flowed through him but was informed by every song he heard and every bit of prose he read. It was his tributary to the river his brilliance swam in. Countless times I’ve heard musicians talk about how a song arrived fully formed from a place they can’t describe (most times, with longing to touch that place again). That place is the river – the continuum of human collective creativity.

As long as we believe in the illusion that creativity is a solitary practice that flows from individual minds, we will misunderstand what LLMs are, what collaboration means, and what it means to be human in an age of thinking machines.

Who wrote "Like a Rolling Stone"?

Dylan held the pencil. But the twenty pages that became the song did not arrive from nowhere. He had spent the previous four years absorbing, with the intensity of a person whose nervous system was calibrated to receive, an extraordinary range of influences. Woody Guthrie's dust-bowl poetry. Robert Johnson's blues compression, entire lives distilled into three chords and a lyric that cut. The Beat poets who gave him permission to break rules he had barely learned. The British Invasion, which was itself an American export reimported with a different accent and a different energy.

Then the England tour broke something open. Exhaustion stripped away the filter. The rant poured out.

Remove any one of those inputs, and the song does not exist. Not a different version. The song itself does not exist, because the song was an act of synthesis, an intelligence shaped by everything before it, every conversation and every record listened to at three in the morning and every argument in a Greenwich Village coffee shop. The sum of those moments found expression in the lyrics, the melody, the beat.

Dylan was not the spring. He was a stretch of rapids in a river that had been flowing long before him, through Guthrie and Johnson and the Delta blues and the field hollers and the work songs and the African rhythms that crossed the Atlantic in the holds of slave ships and the European ballad traditions that came from the other direction. The river ran through him. And because his specific geography – his talent, his timing, his appetites, his location at the confluence of multiple cultural tributaries – created the conditions for a particular kind of turbulence, something remarkable emerged.

But it was not created from nothing. Nothing ever is.

The technical term here is inference. Dylan was taking a vast, implicit training set of cultural experience and producing an output consistent with that training set but not contained within it. The inputs were not new, but the output was. And this reality does not diminish the output. It just describes creative genius more precisely, more diagnostically.

The genius, then, is the person whose particular configuration of inputs, processed through a particular biographical architecture, produces a synthesis that no other configuration could have produced.

But the corollary to that is this: A large language model follows the same operational approach. It just does so with different context and at a different scale.

As offensive as it might be to the traditional creative process, an LLM’s methodology runs nearly parallel to ours. To generate an output, it too performs a structurally analogous operation, a transformation of inputs, experience, and context into something new. There is even a core setting in generating output from LLMs called the "temperature" that governs how much randomness the model allows itself when choosing what to say next. Low temperature: the model takes the safest, most predictable path. High temperature: it reaches for less probable words, less obvious constructions, less expected turns. Call it “creativity” if you want – the engineers do, with the same skeptical quotes – though what it actually means is that the model is less anchored to what it already knows.

This is also why hallucinations don’t disappear when you lower the temperature. LLMs are non-deterministic systems. There isn’t a single correct output, much like how the human mind works.

At their core, hallucinations are a confidence problem. The model generates responses based on the closest probabilistic match in its training distribution, regardless of temperature. Nothing in its architecture forces it to distinguish between what it truly “knows” and what it is merely pattern-matching toward. It simply lands on the most likely answer it can construct.

Grounding mechanisms help. Retrieval-augmented systems help more. But these address the knowledge problem, not the temperature problem. The two are separate dials.

Still, temperature is the interesting one. It's the setting that governs how far the model is willing to stray from the expected. Turn it up, and the outputs get stranger, more surprising, occasionally brilliant, occasionally incoherent. Like the machine getting stoned. The strength of the trip depends on how much you crank it up. You can think of it as dialing up the creativity.

I’ll back up just a bit. Claude is not Dylan. Claude is not creative in the way humans are creative. These are obvious points, but I feel the need to say them because otherwise the argument gets lost in the shouting about who we’re comparing what to. The point is, the distinction we draw between human creation and machine recombination is less stable than we think, because humans also recombine.

We live with the illusion that the people outside our fishbowls are separate from us. We cling to individuality and the romantic idea of creativity in solitude. In reality, we are a social species. Nodes in a larger system. More like a colony than a collection of isolated minds, each of us playing a role in something bigger.

And in humans, the “fittest” is not the individual. It is the collective. It unfolds in concentric circles, from family, to tribe, to the world. We have always created together. Now, that collaboration is massively amplified.

Latent Cultural Space, Illustration by Edo Segal. Ink markers on paper

The recombination is more complex, more biographically specific, more emotionally charged. But the fundamental operation is the same: synthesis from a vast implicit training set through an architecture of its own into something that could not have been predicted. One is done via biological neurons and one by matrix calculations in a silicon based deep learning trained neural network.

The strongest counterargument goes like this: "Sure, Dylan had influences. Everyone has influences. But the synthesis was uniquely his. That is what genius means."

I agree. That is what genius means. And that is exactly my point.

Genius is the quality of the inference, not its independence from a training set. And the same is true of you. Not because you are Dylan, but because you occupy a position in the network that no one else occupies, and the synthesis you produce from your specific set of inputs is, by definition, yours.

If creativity were truly a private reservoir, if the genius really did produce from nothing, then AI would be irrelevant to the conversation. But creativity is relational. It lives in connections between things, in the synthesis of inputs through a particular lens.

And AI is the most powerful demonstration of the relational nature of creativity that has ever existed, because it makes visible what was always true:

The raw material of creation is never original. Only the configuration is. AI for you is as the guitar and harmonica was for Dylan. An instrument to channel and magnify your creative vision.

The biggest creative breakthroughs have always happened between minds. Enlightenment Edinburgh, where David Hume and Adam Smith spent twenty-five years in each other's intellectual orbit, meeting in clubs and taverns, exchanging letters, reading each other's drafts, and where Smith absorbed Hume so completely that Hume became legible on nearly every page Smith wrote, even where he most visibly disagreed with Hume. The Homebrew Computer Club in the 1970s, where Steve Wozniak and dozens of others passed around ideas like joints at a party. In both cases, and every other case, the breakthrough was not inside a single mind. It was in a collision. In the cut, to borrow Raanan's language from our Princeton afternoon – the meaning lives in the juxtaposition, not in either image alone.

What has changed is that there is now a new kind of mind available for collision: A system that can hold your ideas, see them through the prism of ideas drawn from the entire history of human thought, find connections you did not see, and present them in a language you can understand and build further on.

The solitary genius was always a myth. Dylan was never alone – not in Woodstock, not anywhere. He was accompanied by every musician he had ever heard, every poet he had ever read, every argument he had ever lost. The room was crowded with influences. Now the room has a new occupant. And it is not there to compete with the genius or replace the genius. It is there to multiply the collisions. To amplify you.

This brings me to a distinction that will become important in every chapter that follows.

A node is a point in a network: a single mind, a single perspective, a single set of experiences. Dylan was a node. You are a node. I am a node. A node has a location, a shape, a set of connections. Its value is determined not by its independence from the network but by the quality and range of its connections within it.

A network is the pattern of connections between nodes. Intelligence lives there, not in any single node but in the relationships between them. Intelligence is emergent. It arises from connection. We are a hive mind, and LLMs are the first empirical instrument to gaze into that phenomenon.

The Romantic myth says the node is everything. Protect it. Celebrate it. Guard its independence. It is the source of motivation that drives human achievement. The relational view says the node matters because it occupies an irreplaceable location in the network. Its value is its specificity, the particular angle of vision that only this biography, this set of experiences, this configuration of influences can produce. But that value is realized only in connection.

Dylan alone in a vacuum produces nothing. Dylan at the confluence of a dozen cultural tributaries produces "Like a Rolling Stone."

If intelligence is relational, if it lives in the connections between things rather than inside things, then we need a much larger frame than "human vs. machine" to understand what is happening.

We need a frame that goes back to the beginning.

To the first hydrogen atom that found a pattern.