So Daniel's latest prompt lands in the middle of an experiment he's actually running on himself. He's been trying to get off the tail end of a sleep medication while taking Vyvanse during the day, and he's landed on an analogy that's been working for him: the stimulant is the accelerator, the sleep medication is the brake. Push the gas too hard and you get agitation and that skin-crawling overstimulated feeling. Don't push it enough and you can't get going. The trick, he says, is gentle acceleration, gentle deceleration. And with his doctor's approval, he's been splitting his Vyvanse dose in half, four hours apart, using water titration. On a day when he couldn't do that, moving apartments, couldn't find the water bottle, he took it as a single pill and immediately felt the difference. Harder stimulant effect, harder crash. His words: instead of driving calmly, it's an erratic taxi driver slamming the gas and then slamming the brake.
And then he takes the same analogy and points it at alcohol withdrawal, which is where it gets dangerous. He's got alcoholism in his family, his late father died from the disease, and he's clearly done the reading. Glutamate receptors upregulate, the accelerator's stuck down, and if doctors just keep pounding sedatives at the same rate the alcohol was going, the patient never actually withdraws. So the work of detox becomes calculating a descent that minimizes discomfort and eliminates seizure risk. Which is a very clean way of describing a very messy clinical problem.
His actual question, though, is bigger than either the Vyvanse or the alcohol. He wants to know whether doctors have built models that track or simulate what's happening to key neurotransmitter systems as doses change. Not just plasma levels, but the actual signaling. And he asks it in a very specific way: if he walked into a doctor's office tomorrow with an AI model and said, here's how I'm trying to get off this sleep medication without horrible insomnia, and if neurotransmitter levels were easy to measure and display on a dashboard, how useful would that actually be? Could we calculate a titration that lands as smoothly as possible? How mature is our ability to build protocols from real brain signaling rather than from symptoms and guesswork?
That's the question underneath the question. He's not really asking about his own taper. He's asking whether the thing he's doing with a water bottle and subjective feel is a homemade version of something medicine should be able to do properly with data.
Right. He's built a crude titration tool out of a measuring cup and his own attention to how his body feels, and he's wondering where the real version is.
So let's start with the analogy itself, because it's good but it's not quite right, and the place it's not right is exactly where the interesting stuff lives. The brain doesn't have a separate accelerator system and a separate brake system. Stimulants and sedatives act on overlapping circuits. Dopamine and norepinephrine are involved in arousal and attention, but they also talk to the sleep-wake system. GABA, the main inhibitory neurotransmitter, is what sleep medications and alcohol both poke at, but it's also involved in anxiety, muscle tone, seizure threshold. So when Daniel's pressing the accelerator in the morning and the brake at night, he's not operating two independent pedals. He's adjusting two different inputs to a single, very tangled control system.
It's more like one of those old airplane throttles where you've got two levers but they're mechanically linked, and pulling one changes how the other one feels.
That's the image. So why does split dosing feel smoother? Vyvanse is lisdexamfetamine. It's a prodrug, which means the molecule itself is inactive. It has to be metabolized, cleaved by enzymes in red blood cells, to release dextroamphetamine. That's the active drug. And because that conversion takes time, Vyvanse already has a slower onset than immediate-release amphetamine. The manufacturer's whole pitch is that it's harder to abuse because you can't snort it for a rush. But Daniel's finding that even that built-in delay isn't smooth enough for him.
So he's smoothing the smoother.
And the reason it works is pharmacokinetic. When you take the full dose as one pill, you get a plasma concentration curve that rises to a peak and then falls. The peak is where you feel the strongest stimulant effect, and the rate of rise matters as much as the absolute height. Agitation, jitteriness, that overstimulated feeling, those correlate with how fast the concentration is climbing and how high it gets, not just with how much drug you took total. Split the same dose into two halves four hours apart, and you get two smaller peaks. The first one is lower, and by the time the second one arrives, the first is declining but not gone, so the trough between them is shallower. The overall curve looks less like a mountain and more like rolling hills.
A lower peak and a shallower trough. Same total daily exposure, different shape.
Right. And the shape is what his brain notices. The single-pill day, that's a higher, sharper peak. He feels the stimulant push harder, and then when the concentration drops off the back side of that peak, the decline is steeper, and that's the crash. The erratic taxi driver is a pretty good description of a sharp peak-to-trough cycle at the dopamine transporter. The drug blocks reuptake, dopamine hangs around in the synapse longer, and when the drug clears, the transporter comes back online and starts vacuuming dopamine back up. If the concentration drops fast, that vacuum effect is more noticeable. It feels like the floor falling out.
And the split dosing approximates a smoother occupancy curve at the transporter. The drug's coming and going more gradually, so the brain's compensatory mechanisms aren't getting yanked around as hard.
That's the piece about downregulation. When you flood the synapse with dopamine repeatedly, the brain starts pulling receptors back, reducing sensitivity. A sharp spike followed by a sharp drop is exactly the pattern that tends to trigger that compensation. A smoother curve doesn't provoke the same response. It's the difference between a thermostat that cycles on and off in big swings and one that holds a steady temperature.
Which brings up steady state. Daniel mentioned it, the feeling of his brain coming online as the drug reaches steady state in his bloodstream. What's actually happening there?
Steady state is when the amount of drug going in each day equals the amount being cleared, so the plasma concentration stabilizes within a range rather than swinging wildly between doses. For most drugs, that takes about four to five half-lives. Vyvanse's active drug has a half-life on the order of ten to twelve hours, so we're talking several days before the levels really stabilize. And the subjective experience of the brain coming online, that's partly the drug reaching a stable concentration, and partly the brain adapting to having a consistent level of dopamine and norepinephrine signaling. The system stops reacting to each dose as an event and starts treating it as background conditions.
But here's the limitation I want to flag before we move on, because it's the exact gap Daniel's dashboard question is pointing at. Everything you just described, the peaks, the troughs, the occupancy curves, we're inferring it. Nobody measured dopamine in Daniel's synapse. We're reasoning backward from plasma concentrations and his subjective report. Plasma levels of dextroamphetamine are a proxy for what's happening at the receptor. They're a decent proxy, but they're not the thing itself.
That's the honest truth. We have pharmacokinetic models that can predict plasma levels pretty well. We have pharmacodynamic models that try to link those plasma levels to effects, like symptom scores or cognitive performance. But the actual neurotransmitter concentrations in specific brain regions, in real time, in a living human, we can't measure that. Not non-invasively. So everything we say about what's happening at the dopamine transporter during a crash is a reconstruction. Educated, mechanistically grounded, but a reconstruction.
So that's the stimulant side. Now flip it. What happens when the brake is the thing you're trying to remove?
Alcohol withdrawal is the extreme case, and it's worth being precise about the neurobiology because it explains why it's so dangerous. Chronic alcohol use does two things. It downregulates GABA-A receptors. GABA is the main inhibitory neurotransmitter, the brake. Alcohol binds to GABA-A receptors and enhances their effect, so if you're drinking heavily every day, the brain says, I've got too much inhibition, I'm going to make fewer GABA receptors or make the existing ones less sensitive. At the same time, alcohol blocks NMDA receptors, which are glutamate receptors, the main excitatory system, the accelerator. Block the accelerator long enough and the brain compensates by upregulating glutamate receptors, making them more sensitive. So a heavy drinker's brain has a weak brake and a touchy accelerator.
And then you take the alcohol away.
Then you take the alcohol away, and you've got a brain with too few effective GABA receptors and too many sensitive glutamate receptors. The brake is gone and the accelerator is stuck down. That's the hyperexcitable state. It starts with anxiety, tremor, sweating, racing heart, insomnia. But it can escalate to seizures, and in the worst cases to delirium tremens, which is the full-blown medical emergency. Confusion, hallucinations, autonomic instability, blood pressure swinging wildly, fever. Delirium tremens has a mortality rate that's not trivial even with treatment. This is why detox is a medical procedure. It's not willpower. It's keeping someone alive while their brain re-regulates.
And Daniel's point about not just pounding the patient with sedatives forever, that's the actual clinical tension. You need to give enough benzodiazepine to keep the GABA system from collapsing into seizures, but you need to be reducing it over time so the brain has to rebuild its own brakes.
That's exactly the art of it. So how do doctors actually do this? The standard tool is a scale called the CIWA-Ar, the Clinical Institute Withdrawal Assessment for Alcohol, revised. It's a scored assessment. A nurse or doctor asks the patient a series of questions and observes them. Nausea and vomiting, tremor, sweating, anxiety, agitation, headache, orientation, and a few specific ones. Tactile disturbances, which is the question about whether it feels like bugs are crawling on the skin. Auditory disturbances, visual disturbances. Each gets a score, and the total tells you how severe the withdrawal is right now.
And the benzodiazepine dose is tied to that score.
Two main approaches. Fixed schedule, where you give a set dose at set intervals and taper it down over several days. Or symptom-triggered, where you only give the drug when the CIWA score crosses a threshold, and the dose is proportional to the score. Symptom-triggered dosing tends to use less total benzodiazepine and shortens the detox, because you're not giving the drug when the patient doesn't need it. But it requires staff who can assess frequently and reliably.
So the protocol is a hand-tuned titration curve. The clinician is watching the score, adjusting the brake in real time, trying to keep the patient in that zone where the glutamate storm is suppressed enough to prevent seizures but the GABA system is still being forced to recover.
And here's the thing that connects back to Daniel's question. The CIWA score is not a neurotransmitter measurement. It's a behavioral and subjective proxy. The nurse isn't measuring GABA levels. She's asking whether the walls are moving and writing down a number. The entire state of the art in acute alcohol withdrawal is a trained human looking at symptoms and guessing the dose. It works, it saves lives, but it's not a dashboard.
So that's the acute context. Now bring it back to Daniel's situation, the chronic outpatient taper. Getting off the end of a sleep medication. Same problem, lower stakes.
Lower acute stakes, but the same underlying dynamic. If it's a benzodiazepine or a Z-drug, you're dealing with GABA-A again. Taper too fast and you get rebound insomnia, anxiety, potentially seizures if it's a high enough dose of a benzo. If it's an orexin antagonist, which is the newer class, the suvorexant and lemborexant type drugs, the mechanism is different. Orexin is a wake-promoting neuropeptide, and blocking it promotes sleep. Tapering that is less about seizure risk and more about the insomnia coming back as the receptors re-sensitize to orexin. But the clinical problem is the same: how fast can you descend the dose curve without triggering the symptoms you're trying to avoid? And the answer right now is, we guess, we go slow, we adjust based on how the patient reports sleeping.
Which is Daniel standing in his kitchen with a water bottle, splitting a Vyvanse capsule, because he's noticed that the shape of the curve matters more than the total dose.
Right. He's doing manually what a proper model would do computationally. So now we can actually answer his question. Have doctors developed models that track or simulate effects on key neurotransmitter systems as doses change? The answer is partial, and the partial part is important. Pharmacometrics, which is the field of building mathematical models of drug behavior, can simulate plasma concentrations very well. We have something called PK/PD modeling. PK is pharmacokinetics, what the body does to the drug. PD is pharmacodynamics, what the drug does to the body. A PK/PD model can take a dosing schedule and predict the plasma concentration over time, and then link that concentration to a measured effect, like a symptom score or a physiological marker.
So you could model Daniel's split-dose Vyvanse and show the lower peak and shallower trough, and maybe even link that to a predicted side-effect profile.
You could. Model-informed precision dosing is the term that's been picking up steam. You take a patient's weight, age, liver function, kidney function, maybe a genetic test for metabolic enzymes, and you build a model that predicts how they'll clear the drug. Then you simulate different dosing schedules and pick the one that gives you the curve you want. This is already used in some areas, antibiotics in critically ill patients, certain cancer drugs, immunosuppressants. But the effect side, the PD side, is almost always a proxy. Plasma concentration is the thing you measure, and you assume the effect tracks the concentration.
So the model stops at the blood-brain barrier, more or less.
That's the cleanest way to put it. The model tells you how much drug is in the blood. It does not tell you what's happening to dopamine in the prefrontal cortex or GABA in the amygdala. It can't, because we can't measure those things in real time in a living person. There are research tools, microdialysis, PET imaging, but microdialysis involves sticking a probe into brain tissue, and PET requires injecting a radioactive tracer and lying in a scanner. Neither is something you can wear around the house while you're tapering a sleep medication.
What about the AI angle? Daniel's asking about walking into a doctor's office with an AI model. Is there anything real there?
There's something real, and it's worth being precise about what it is and isn't. There's been a wave of work in the last couple of years on using machine learning for what's being called neurotransmitter modulation, but the modulation is indirect. The AI isn't measuring dopamine. It's taking in proxy data, plasma levels, symptom questionnaires, wearable data like heart rate variability and sleep architecture, maybe actigraphy, and it's learning patterns that predict how a patient will respond to a dose change. Then it suggests a schedule. That's model-informed precision dosing with a machine learning layer. It's useful, it can catch patterns a human would miss, but it's still operating on proxies.
The AI is a better guesser, not a measurer.
That distinction matters because it's the answer to the dashboard question. Daniel asked, in an imaginary world where neurotransmitter levels were easy to measure and display, how useful would it be to calculate a titration that landed as smoothly as possible? The answer is, enormously useful, and we can say that with confidence because we know what the dashboard would need to show and which pieces we're missing.
Walk through the pieces.
Three components. First, real-time measurement. You need a sensor that can tell you dopamine concentration in the relevant brain region, moment to moment, non-invasively. That doesn't exist. Not close. There's work on electrochemical sensors, there's work on wearable spectroscopy, but the brain is encased in bone, and the molecules are present in tiny quantities in specific regions. You can't just shine a light through the skull and read dopamine. The signal-to-noise problem is enormous.
Second piece.
A model linking neurotransmitter levels to symptoms. Even if you could measure dopamine in real time, you'd need to know what a given level means. What concentration produces focus, what concentration produces agitation, how that threshold varies by person and by time of day and by what else is in the system. That's a massive data problem. We have fragments of this from animal studies and from PET imaging, but nothing like a complete map.
And third.
A control algorithm. Something that takes the measured level, compares it to a target, and computes the next dose adjustment. That part, honestly, is the easiest. Control theory is mature. We use exactly this kind of algorithm in insulin pumps. Continuous glucose monitor measures glucose, algorithm computes insulin dose, pump delivers it. That's a closed-loop system for a single molecule that we can measure continuously. The glucose sensor is the thing that makes it work. Without the sensor, the algorithm has nothing to steer by.
The bottleneck is measurement, not modeling. That's the line I want to sit with.
It's the line. The modeling is hard but it's advancing. The measurement is the wall. And it's not just a technology problem, it's a biology problem. Neurotransmitters act in synapses, in tiny volumes, at micromolar concentrations, on timescales of milliseconds. Even if you could sample the extracellular fluid in the prefrontal cortex, you'd be measuring a spillover signal, not the actual synaptic signaling. And the relationship between extracellular concentration and receptor activation is nonlinear and depends on receptor density, which is itself changing in response to the drug. So the dashboard Daniel's imagining, with a clean number for dopamine and a clean number for GABA, would be a simplification even if we could build it.
But here's the thing. Simplifications are useful. His accelerator-and-brake analogy is a simplification, and it's been useful to him. The CIWA score is a simplification, and it saves lives. So the right question isn't whether a neurotransmitter dashboard would be perfectly accurate. It's whether a rough version would be better than what we have now.
The answer is probably yes, for a specific subset of problems. The taper problem, the titration problem, the problem of finding the smoothest path from one dose to another, that's exactly the kind of problem where even a noisy signal would help. Because right now we're doing open-loop control. The doctor prescribes a taper schedule, the patient goes home, and the feedback arrives days or weeks later in the form of a phone call or an appointment. The loop is slow. Any real-time signal, even a proxy, would tighten that loop dramatically.
We're not as far as the dashboard, but we're not nowhere. The proxies are getting better.
Wearables are the interesting frontier. Heart rate variability tracks autonomic arousal, which is downstream of the norepinephrine system. Sleep architecture, measured by a decent wearable, tells you something about GABA and orexin function. Symptom diaries, if done carefully, capture the subjective experience that we ultimately care about. None of these are neurotransmitter levels, but they're all correlated with the systems we're trying to steer. And the machine learning work is getting better at combining them into a picture.
Daniel's homemade version, the water titration and the attention to how he feels, is actually a pretty good proxy dashboard. He's measuring subjective state continuously, he's adjusting the dose curve manually, he's doing closed-loop control with a human in the loop.
That's the thing I keep coming back to. The most sophisticated titration system in this story is Daniel with a water bottle and a careful attention to his own body. Not because he's a genius, but because he's the only sensor that exists. He's the only thing that can measure how he feels, and he's using that measurement to adjust the curve. The entire field of pharmacology is trying to build an external version of what he's doing internally.
Where does that leave the question of how mature our ability is to create protocols from real brain signaling? I'd say the honest answer is, the protocols exist, the signaling we're reading is a proxy, and the gap between the two is the entire field.
That's a fair summary. The protocols are real. Symptom-triggered benzodiazepine dosing in alcohol withdrawal is a protocol that adjusts in real time to a measured signal. The signal is just a nurse with a clipboard, not a sensor. And the chronic tapering protocols, the slow descents that doctors prescribe, they're built on decades of clinical experience about what works and what causes rebound. They're not built on neurotransmitter models. They're built on outcomes.
Which is a different kind of evidence, and not a worse one. It's just not what Daniel's asking about. He's asking about the mechanistic layer, the actual signaling, and the answer is that layer is still dark to us.
It's worth saying plainly, because there's a misconception floating around that we already have the technology and we're just not using it. That's not true. We cannot measure neurotransmitter levels in real time in a living human brain non-invasively. Not with anything on the horizon that I've seen. The bottleneck is fundamental.
That's the point where I want to bring in someone who's actually seen the crude version of this dashboard in action.
Hilbert: The dashboard already exists. It's a piece of paper with a list of questions and a number at the bottom. I worked the night shift at a detox facility in ninety-eight, intake coordinator. I logged the CIWA scores into the chart. Same patients came back three, four times. One guy, his seizure started in the parking lot, before he even got through the door. The nurse said we're chasing the brake pedal. She didn't mean it as a metaphor. She meant the dose of Ativan she was about to push.
Chasing the brake pedal. That's the symptom-triggered protocol in one line. You're always behind the withdrawal, trying to catch up.
Hilbert: That's what it was. The score tells you where the patient was ten minutes ago. The seizure tells you where he is now. The gap between those two things is where the danger lives. So when Daniel says, what if we had a dashboard, I'd say we've got one. It's just slow. It's a human being asking whether the walls are moving and writing down a number. The gap between that and a real-time brain readout is not a technology gap. It's a measurement gap. And I'm not sure we close it in my lifetime.
The score sheet is the dashboard. That's the point. It's a proxy, it's delayed, it's subjective, and it's still the best tool we have.
Hilbert: I kept one of the old sheets. It's in a drawer. The questions are printed on it. One of them says, patient reports feeling like bugs are crawling on skin. That's the whole thing, right there. That's what a neurotransmitter dashboard looks like in the real world. A person describing a symptom, and a clinician guessing the dose. The glutamate storm is happening inside the skull, and the best readout we've got is a man saying he feels bugs.
The tactile disturbance item on the CIWA, that's not a random question. It's a specific symptom of the hyperexcitable state. The sensory cortex is firing without input, and the brain interprets it as bugs. It's a direct window into the glutamate overdrive, but it's a window through frosted glass.
Hilbert: I logged that score at three in the morning more times than I can count. The nurse would look at the number, look at the patient, and pick a dose. Sometimes she'd wait and re-score in an hour. Sometimes she'd push the Ativan right then. She was doing what Daniel's doing with his water bottle, just with higher stakes and a shorter clock.
The same fundamental limitation. She was reading a proxy and adjusting a curve. She wasn't measuring GABA. She was measuring what a man said about his own skin.
Hilbert: The thing about the dashboard question is, Daniel's asking whether we could do better with a real-time readout. And the answer is, obviously yes. If you could watch the glutamate climb in real time, you'd dose before the seizure instead of after it. You'd stay ahead of the curve instead of chasing it. But the readout doesn't exist. And the reason it doesn't exist is not that nobody's tried. It's that the brain doesn't want to be measured. It's bone and fluid and tiny amounts of chemical in tiny spaces, changing faster than any sensor we've got can track.
The timescale point is underrated. A seizure can develop in minutes. A panic attack, seconds. The dopamine shift that causes a stimulant crash, that's happening over an hour or two. Any sensor that samples every five minutes is missing most of the action. Glucose monitors work because glucose changes slowly, over tens of minutes. Neurotransmitters don't cooperate like that.
Hilbert: My brother-in-law works in a lab that does microdialysis in rats. He says the probe is the size of a needle and it measures one tiny spot, and the rats have to be anesthetized or at least tethered. He says the idea of doing that in a person walking around living their life is science fiction. And he's not a guy who uses the phrase science fiction lightly. He's also not a guy I'd trust on anything involving money. But on the probe, he knows what he's talking about.
The pieces exist in isolation. The measurement tools exist but they're invasive and slow. The models exist but they're built on proxies. The control algorithms exist but they need a signal to steer by. And the integration, the thing Daniel's imagining, doesn't exist because the middle piece is missing.
That's the summary. And it's worth saying that the integration is where the field is heading, slowly. The closed-loop insulin pump is the proof of concept. The question is whether we can find a proxy signal that's good enough to close the loop for brain chemistry. Heart rate variability, skin conductance, sleep metrics, maybe some combination. It won't be a neurotransmitter dashboard. But it might be good enough to make tapers smoother.
Hilbert: The nurse with the clipboard was good enough to keep most people alive. Most. And she was working with a delay of minutes and a signal that was a man describing bugs. If you could cut that delay to seconds, even with the same signal, you'd save lives. That's the part that gets lost in the talk about real-time measurement. You don't need perfection. You need faster.
That's the open question I want to leave hanging. If we can't measure neurotransmitters directly, what's the best proxy we can build? What combination of wearables and symptom tracking and maybe some blood markers gets us close enough to make the titration modeling actually useful?
The gap between pharmacometrics, which models plasma concentration, and psychiatry, which treats symptoms, that's where the next generation of personalized dosing tools has to live. It's a gap that's been there for decades, and the only thing that's changed is that the proxies are getting denser and the machine learning is getting better at finding patterns in them. But the fundamental measurement problem hasn't moved.
Daniel's split-dosing experiment is a small, personal version of this whole problem. He's using subjective report and a water bottle to approximate a smoother curve. The question is whether medicine can do that with data instead of guesswork. And the answer, right now, is that medicine is doing it with a better class of guesswork, but it's still guesswork. The dashboard is a long way off.
The thing that should stick, the one idea from this whole discussion, is that the bottleneck is measurement, not modeling. We know how to build the algorithms. We know how to simulate the curves. We don't know how to read the actual signal. And until we do, every titration, from Daniel's water bottle to the CIWA score sheet, is a human being interpreting a proxy and adjusting a curve by feel.
The dashboard Daniel's imagining would be transformative. But the closest thing we have to it right now is a man noticing how his own brain feels and a nurse asking whether the walls are moving. The gap between those two things is the entire field.
That's the open question to sit with. What proxy signal, what combination of wearable data and subjective report and maybe something we haven't thought of yet, gets us close enough to make the modeling useful? Because the modeling's ready. It's waiting for something to measure.
Thanks to Hilbert Flumingtop for producing, as always.
This has been My Weird Prompts. If you want to send us your own strange question, email us at show at my weird prompts dot com.
We'll be back soon.
