Your Brain Generates Waves That Predict What You See Next
Here's something that should mess with your head: your brain generates actual traveling waves. Not the metaphorical kind you hear about in meditation apps, but coordinated sweeps of electrical activity that propagate across the surface of your cerebral cortex like waves moving through shallow water.
These neural traveling waves (nTWs) were first spotted in anesthetized animals back in the day, but Salk Institute neuroscientist John Reynolds and his team found them everywhere in awake animals too. And here's the kicker—they directly correlate with whether you actually perceive something right in front of your face. That pair of keys you're sure you looked right at? The wave phase passing through your visual cortex at that exact moment determined whether you saw them or not.
Reynolds first identified these waves in awake visual systems back in 2020. His lab found that an animal's ability to perceive a visual object depended directly on where and when a wave was passing through its visual cortex at that moment. It explained the classic conundrum of searching for something right in front of you that somehow goes completely unnoticed. You were looking at it. Your brain just wasn't letting the signal through because the wave was in the wrong phase.
But Reynolds kept pushing. He wanted to know why these waves exist at all. What are they actually computing? The answer, published in Neuron on July 21, 2026, is that they're doing something functionally analogous to large language models—building a generative model of the world from experience.
The Four Things These Waves Do
Reynolds' paper lays out a framework that's honestly kind of mind-blowing. The visual cortex uses traveling waves to do four things simultaneously, all implemented by the same underlying wave dynamics:
Modulate perception moment to moment. Your brain isn't just passively receiving visual input. These waves actively shape what you see right now, when you see it, and whether that stimulus actually makes it into your conscious awareness. The instantaneous phase of a passing wave determines perceptual sensitivity at any given cortical location.
Build sensory representations. Recent sensory inputs get transformed into internal models. Your brain isn't storing raw pixels—it's constructing something that represents what those inputs mean in the context of your environment.
Generate short-term predictions. The waves predict what's coming next. Not in a psychic way, but through learned statistical regularities of the physical world—how objects move, how light behaves, what physics governs your surroundings.
Replay temporal memories. Events that unfold over time get stored as patterns the brain can replay. Think of it like your visual cortex keeping a highlight reel of recent experiences, ready to be reactivated when similar patterns emerge.
What makes this framework so powerful is that these aren't separate processes running in isolation. They're all implemented by the same underlying wave dynamics—temporal wave patterns that introduce spatiotemporal dependencies across sensory maps that you simply can't capture with feedforward or feedback processing alone.
Your Brain Is Basically an LLM
Here's where it gets weird. Reynolds draws a direct parallel between neural traveling waves and large language models like ChatGPT.
Think about what an LLM does. It learns statistical structure from language—patterns of how words co-occur, how sentences unfold, what comes next in a paragraph. Then it uses that learned structure to generate text that reflects those patterns. The model doesn't understand meaning the way you do, but it captures something functionally similar: regularities in the data.
Your brain does the same thing with visual experience. Recurrent cortical wave circuits learn the physical and statistical regularities of your environment—the 3D structure of objects, how light behaves, how things move according to physics. Then traveling waves use those learned synaptic patterns to fill in missing details, predict upcoming visual changes, and generate an internal model of the physical world.
"This is, in a meaningful sense, analogous to what large language models like ChatGPT do," Reynolds explains. "They learn statistical structure from language and use that knowledge to generate meaningful and appropriately structured text that reflects the patterns of language. The brain may be doing something functionally similar—a biological generative model built from the ground up by experience."
The analogy isn't perfect—brains are wet, messy, and embarrassingly biological—but the computational principle is strikingly similar. Both systems internalize statistical structure and use it to generate predictions. The difference is scale and substrate, not principle.
The Synaptic Weight Secret
Here's what really separates this from a simple feedforward relay: the neural connections that generate these waves aren't just passing signals along. They're actively changing.
Every sight, smell, sound, and action you experience alters the connections that generate these waves. The synaptic weights—the physiological properties of how neurons communicate—adapt based on incoming sensory feedback. Your brain is constantly rewriting its own wiring based on what it encounters.
This isn't some slow, developmental process either. It's happening in real time, moment to moment. Each sensory experience contributes to building the neural circuitry your brain uses to construct an internal representation of the external world. The recurrent neural connections that generate traveling waves actively adapt their physiological synaptic weights based on incoming sensory feedback.
Think about that for a second. The same recurrent circuits that generate traveling waves are also the circuits that get modified by experience. The output shapes the input, which reshapes the output, which... you get the picture. It's a loop that never stops.
This dynamic synaptic weighting is what makes the whole framework work. Without it, you'd just have a passive relay system—signals going in one end and coming out the other. But with adaptive weights, the waves themselves become a learning mechanism. The brain isn't just processing the world. It's continuously updating its model of the world through the very waves it generates.
From Noise to Engine
Historically, rhythmic cortical oscillations were dismissed as byproduct "noise" or treated as simple passive global clocks. Just background static, basically.
Reframing traveling waves as active computational engines changes everything. It demonstrates that temporal wave dynamics actively perform complex calculations—allowing the brain to transform messy sensory inputs into structured, predictive perceptions with minimal energy cost.
This isn't just semantics. It's a fundamental shift in how we understand brain function. Those waves aren't happening despite the computation—they are the computation.
The implications are huge. If traveling waves are doing this kind of heavy lifting, then disruptions to wave dynamics could explain a whole range of perceptual and cognitive disorders. And if we can learn to modulate these waves, the therapeutic possibilities become almost scary.
What's particularly elegant is how this reframing resolves a long-standing puzzle in neuroscience. For decades, researchers observed these rhythmic patterns but couldn't figure out what computational job they were doing. Were they just epiphenomena—side effects of other processes? Or were they actually doing something useful?
The answer, apparently, is the latter. These waves are computational engines, not byproducts. And that changes how we think about everything from attention to consciousness to the very nature of perception.
Inferring Causes From Chaos
Here's the thing about sensory input: it's messy. Noisy. Incomplete.
Your brain receives retinal images that change constantly with your eye and body movements. Objects exist in 3D space, but you're seeing them from a 2D projection that shifts every time you move. The laws of physics and physiology overlay all of this, but the raw data your brain gets is gloriously chaotic.
So what does your brain do? It internalizes the regularities. The 3D spatial structure of objects. How physics governs motion. How your body and eyes move through space. All of this gets encoded in networks of synapses.
Then those synapses generate waves that allow your brain to infer the likely causes of sensory input. You're not just seeing what's in front of you—you're actively constructing a model of what that input means.
This is predictive inference in its purest form. Your brain isn't waiting for perfect data. It's making educated guesses about what's causing the sensory input it receives, then using those guesses to guide perception and behavior.
The traveling waves are the mechanism that makes this possible. They embed sensory history in the evolving activity pattern of cortical regions, creating a continuous stream of prediction and correction. Each time the brain receives sensory input, it has to decide: what am I most likely sensing right now?
The world is rich and complex, but it's also somewhat predictable. Objects around you exist in 3D space. You see the world in retinal images that change with your eye and body movements. The laws of physics and physiology overlay all of this. Reynolds' framework proposes that the brain internalizes these regularities by encoding them in networks of synapses, which then generate waves that allow inference from noisy sensory input.
This framing brings us one step closer to understanding how our brains compute the busy, messy world around us—turning a complicated sensory onslaught into coherent behaviors and experiences.
For a related look at how predictive processing shapes visual experience when sensory input fails, see The Neural Architecture of Phantom Vision.
The Circuit Mechanisms
What makes these waves possible are recurrent cortical circuits—neural networks where neurons loop back on themselves, creating the conditions for wave propagation. These aren't simple one-way highways. They're complex feedback loops that can sustain and propagate electrical activity across cortical surfaces.
The waves can arise intrinsically in ongoing network activity, or they can be triggered by sensory input and behavioral events. This dual origin matters because it means the brain isn't just reacting to the world—it's also generating its own internal dynamics that shape how those reactions unfold.
By structuring neural activity within individual cortical regions, nTWs introduce spatiotemporal dependencies across sensory maps that aren't naturally captured by purely feedforward or feedback processing. In other words, the waves create correlations between different parts of your visual field that wouldn't exist if you were just processing information in a straight line from retina to cortex.
These spatiotemporal computations—predicting upcoming sensory inputs by embedding sensory history in evolving activity patterns—are what give traveling waves their computational power. They're not just moving electricity around. They're performing calculations that would be incredibly expensive, or even impossible, with standard feedforward architectures.
Neural oscillations aren't unique to vision. Research on wave dynamics in auditory and musical perception reveals the same principle across sensory systems—see When the Brain Becomes the Instrument: How Neural Resonance Theory Rewrites Music Perception.
Why This Matters
This research doesn't just sit in a journal somewhere. It reframes how we think about perception, prediction, and the very nature of consciousness.
If your brain is essentially running a biological generative model powered by traveling waves, then perception isn't passive reception—it's active construction. You don't see the world as it is. You see the world your brain predicts it should be, corrected by sensory input when predictions fail.
The LLM analogy isn't just a catchy soundbite. It points to a deep computational principle: that learning statistical structure from experience and using that structure to generate predictions is a universal strategy, whether you're made of silicon or neurons.
And the implications for disorders of perception? If traveling wave dynamics are disrupted, the whole predictive machinery breaks down. Schizophrenia, autism, attention disorders—could they all involve broken wave dynamics? The framework opens up entire new avenues for understanding and potentially treating these conditions.
Reynolds' paper, with co-authors Lyle Muller (UT Dallas and Fields Institute), Alexandra Busch (Fields Institute and Western University), and Zachary Davis (University of Utah), represents a genuine synthesis of physiological and computational evidence. It's not just another review article. It's a roadmap for the next decade of neuroscience research into how the brain computes.
Attention itself is shaped by how cortical areas coordinate—for a window into how higher-order regions modulate that coordination, see Inside the Brain's Editing Room: How the Frontal Lobe's Audiovisual Map Coordinates Human Focus.