Sleep runs on two clocks — and worrying about it can be worse than a short night

Lex Fridman opens with the biggest question he can find: why do humans need sleep? Andrew Huberman, a Stanford neuroscientist back for his second conversation on the show, wriggles out of a final answer. On any *why* question, he says, he defaults to the same line: he was not consulted in the design phase.

What he can describe is mechanism. The longer you stay awake, the more adenosine accumulates in the brain, binds to adenosine receptors, and produces sleepiness — a pressure that does not care whether it is day or night. How sleepy a given pile of adenosine actually feels depends on a second, separate system: a highly conserved circadian oscillation that is, at bottom, a temperature rhythm.

If you are awake by day and asleep at night, core temperature usually bottoms out around 3 or 4 a.m., climbs after you wake, peaks in the late afternoon, and falls again toward evening. The whole swing takes about 24 hours, plus or minus an hour, lined up with the Earth's spin. Sleepiness, Huberman says, is adenosine multiplied by where you sit on that curve.

The cleanest way to pull the two systems apart is to stay up all night. Huberman uses himself: by 3 or 4 a.m. he is very sleepy as adenosine keeps rising, but around his usual wake time of 5, 6, or 7 a.m. he starts to feel more alert even though adenosine is higher than it was a few hours earlier. The clock has moved into a more wakeful phase, and that phase can overrule the chemical debt for a while.

Every cell in the body runs its own roughly 24-hour rhythm, driven by genes such as *clock* and *per*. Huberman calls that one of the great successes of biology — work in flies, bacteria, and mammals associated with people like Steve Reppert, Michael Rosbash, and others — but local clocks still need a conductor. The conductor is the suprachiasmatic nucleus, a master circadian clock sitting just above the roof of the mouth.

That master clock does not try to visit every tissue with a custom peptide. It controls systemic temperature, and temperature is what aligns the rest. Joe Takahashi, formerly at Northwestern and later at UT Southwestern, showed that the story coming out of the master clock boils down to making tissues oscillate in temperature together. A peptide-and-receptor scheme, Huberman argues, would leave too much room for binding affinities, cell turnover, and organs at different stages of maturation.

Pull a liver out and set it on a table, and those clock genes still oscillate on their own. They are independent; they just need to be entrained. Temperature is an attractive shared signal because it reaches every tissue the same way. Huberman also notes that strictly nocturnal people, even if they sleep enough to push adenosine back down, do worse on immune and metabolic measures than people who are awake in daylight. Human hardware was tuned for a diurnal schedule. Crepuscular predators such as mountain lions are a different design: rod-cone ratios that let them hunt when prey cannot see, the biological equivalent of night-vision goggles.

Falling asleep is easier when the body is cooling. Huberman says the brain and the rest of the body need to drop about two to three degrees to reach deep sleep; a cold room with a warm blanket or socks can serve that drop without making you too uncomfortable to drift off. Warming later is part of the wake-up cascade: rising temperature helps trigger cortisol from the adrenals. The landmark on that curve is your temperature minimum, typically about two hours before the time you would wake if you slept without an alarm for a few days — not a bathroom trip, not a flight.

That landmark decides what light does to you. Bright light in the eyes in the two to four hours *before* the temperature minimum delays the circadian clock: the whole oscillation slides later, so you want to sleep later and wake later the next day. The same bright light in the hours *after* the minimum advances the clock, so you want to go to bed earlier and wake earlier. For Huberman, a typical 6 a.m. wake puts the minimum near 4 a.m.; light at 5, 6, or 7 a.m. pulls the schedule forward.

Most people still sleep and wake at roughly the same times because both forces run at once. Morning light advances the clock a little; evening light delays it a little; you get captured in the middle, oscillating with the spin of the Earth. Fridman, who texts Huberman at odd hours, is the messy counterexample. Huberman's point is not that a chaotic schedule is impossible. It is that the light-and-temperature rule is how the hardware is actually steered.

Fridman says he tries to minimize stress about rest: if he obsesses over a perfect night, he feels worse from the worry than from the lost sleep. Huberman agrees that a dysregulated schedule and a happy, productive life can coexist, and that much of his own life has included all-nighters and odd hours. He thinks the public campaign for sleep has gone too far. The new problem is meta-stress: the fear that missing sleep will cause dementia or crash the reproductive axis, piled on top of whatever the night actually did.

Fatigue is not only adenosine and circadian phase. Effort releases epinephrine, and when those levels run high enough you feel mentally and physically spent and want to quit. Dopamine — the molecule of pursuit, reward, and feeling good — resets the capacity for effort. Biochemically it is what epinephrine is made from: tyrosine to dopa to dopamine to epinephrine. Enjoying the thing you are doing creates a chemical milieu that supplies more of the molecules that allow work. Huberman says that sounds new-age until you open a biochemistry textbook; the books just skip the happiness language and stop at dopamine.

Near the temperature minimum, he adds, the brain is not trustworthy. Thoughts and feelings in that trough should not be given much weight. A colleague's work at Stanford has even tied positive anticipation of the next day to better sleep quality when sleep is short. And a study from Bob Stickgold's lab at Harvard Medical School, with Emily Hoaglund, looked at highly motivated students in organic chemistry: consistency of total sleep duration predicted exam performance better than total sleep duration itself. People who slept a steady six hours outperformed people who bounced between five and nine.

Fridman presses the analogy with nutrition: science answers population questions, such as which diet reduces obesity, not which diet makes David Goggins the best version of himself. High performers are individuals. Huberman's sharper line is that there is still no evidence eight hours is automatically better than six. Someone telling you that you must sleep eight may be right for them and wrong for you.

Huberman's next parameter is ultradian: waking and sleep are both broken into roughly 90-minute cycles. Attention, math, driving — performance ramps, peaks, and falls inside that window. Sleep walks through stages one through four and REM on the same kind of cycle. Ending sleep near the close of a 90-minute block, say after six hours, is often better than sleeping a seventh hour and being yanked out mid-cycle. Huberman says he functions better on six hours than on seven, even though eight hours *feels* great; he cannot remember an uninterrupted eight-hour night since infancy.

A practical test is how you feel during the day. Almost everyone hits an energy dip in the late afternoon, around the temperature peak. That is a good window for a nap of 90 minutes or less, or, if you cannot nap, for lying down with legs elevated so the lymphatic "sewer system" of the brain can clear waste. Huberman's lab studies what he calls NSDR, non-sleep deep rest: lying down and dropping into real calm, including free hypnosis scripts from his Stanford colleague David Spiegel. At least one Danish study, he says, found that a 20-minute NSDR protocol reset neuromodulators such as dopamine in the basal ganglia to levels seen after a long night of sleep.

Fridman treats short naps as magic regardless of the night before. Huberman's recommendation to friends is 20 to 30 minutes, long enough to help and short enough, if you are not badly sleep-deprived, to stay out of REM. Cross 90 minutes and you can drop into REM, where space and time uncouple; coming back from that state for a meeting is disorienting unless you plan to stay there all night. The exception is severe deprivation: land jet-lagged, hit a hotel bed, and you can fall straight into a psychedelic REM dream, which can be great on its own terms.

Fridman is about to run a 4x4x48 with David Goggins on March 5: four miles every four hours for 48 hours, streamed live. The schedule leaves about three hours off the course each round, minus the time it takes to fall asleep, so two hours of rest is optimistic. Huberman sees two workable approaches. One is to hammer through: keep alertness and adrenaline high and do not expect sleep. If you cannot sleep you will not be stressed about it; if you do, it is a bonus. Friends from the SEAL teams have told him that in Hell Week, the offered nap can feel worse than staying up, because being woken 20 minutes later is its own injury.

The other approach is to harvest the 90-minute cycles: come back from a run, wait out the first wired minutes, and take whatever sleep you can. Food is a competing variable. Anything in the gut draws blood toward digestion and pulls you toward the parasympathetic "rest and digest" state. Fridman overate the last time he tried a long effort and made it miserable; he is considering eating almost nothing. Fasting raises epinephrine, Huberman notes, because the system is built to send a hungry animal out to find food. If ancestors had gotten tired the moment they got hungry, we would not be here.

Goggins, texted in advance, sent his own protocol: be organized, waste no time, compress eating and rehab so you can sleep as much as possible — and then the punch line. The only pill that might help is an SEIU pill, hard to get: Suck It Up. Huberman refuses the word *hack* for the rest of this toolkit. It is disrespectful to actual hackers, and it implies you are tricking a system that was selected to work under variable conditions. Gratitude is not a trick. If it is genuine — attached to a larger goal, or to a relative who has died — it releases serotonin that soothes and dopamine that yields more epinephrine. You cannot pretend your way into that chemistry. Fridman found last time that *feeling* grateful was fuel, but turning on a camera to say it in words was the most draining part of the event.

Sleep deprivation, Huberman says, makes people irritable because REM — which does the emotional processing — occupies more of those 90-minute cycles toward morning. Early-night slow-wave sleep is better for muscular repair and some forms of learning; lose REM for too long and little things start to feel large. The body will compensate if Fridman does fall asleep during the challenge: increasing fatigue drops you into REM faster. Humor, he adds, is not a mood trick. It is dopamine, and dopamine is a fresh perspective in chemical form. Talking while exhausted still costs epinephrine you might rather spend on running.

On the course there is a second-by-second control loop among breath, heart, and brainstem. Inhale, the diaphragm drops, the heart gets a little more room, blood moves more slowly through that larger volume, and neurons around the sinoatrial node signal the brainstem to speed the heart. Exhale, the reverse: the heart gets smaller, flow speeds up, the brainstem slows the heart. That interplay is the basis of heart-rate variability. If your heart is racing and each unit of effort feels too expensive, lengthen or intensify the exhale. If you are truly flagging, bias toward faster inhales to pull in more oxygen and pick the heart rate up.

During steady running Huberman likes a double inhale — a breath plus a top-up — then an exhale. The double inhale reopens tiny sacs in the lungs that collapse as you fatigue and carbon dioxide builds; reinflating them dumps CO₂. Nose for the double inhale, mouth for the offload, until you hit a sprint, at which point he says ditch the pattern and use whatever is there. Over-controlling breath spends epinephrine too. The goal is a state of exertion you do not have to think about.

Fridman asks about anger, because Goggins will push buttons and because Goggins sometimes invents enemies in order to have something to fight. Physiologically, Huberman says, autonomic arousal looks almost identical for love and excitement as for anger and the wish to defeat someone. Love more often comes with serotonin and dopamine that replenish. He will not recommend living in friction, but he admits some of his best work and hardest deadlines came from not wanting to be outcompeted. Anger, channeled, is powerful fuel. Fridman tried an anger-driven challenge for 30 days and found the emotional roller coaster exhausting. Better, he concluded, to use anger as an occasional boost and otherwise get lost in the rhythm of the work.

Huberman takes the anger question into endocrinology. The major effect of testosterone, he says, is to make effort feel good. The molecule is synthesized from cholesterol, and the same cholesterol can be made into cortisol instead. You cannot do both with the same unit. Stay in anger long enough and you divert more of that pool toward stress, slowly depleting the pathway that makes work feel rewarding. Sleep deprivation does not necessarily drop testosterone on the spot — that, he says, is a common myth — but sleep is eventually required to replenish it. The live question is whether you are enjoying what you are doing. Work from Duncan French, including a thesis at UConn, maps the triangle of stress hormones, testosterone, and dopamine. When effort feels good, Huberman calls that, aside from plasticity, the holy grail: not the medal at the end, the process.

He tells a story he has on film. A senior colleague dying of pancreatic cancer said he did not want to make it past the new year. Huberman knew the man was on androgen therapy, asked whether he had used his cream, and watched him apply it. Ten minutes later the colleague said he thought he did want to live into the new year, and that he was going to write 12 letters of recommendation — which he did. Huberman is not telling people to take androgens. He is saying these molecules are anciently tied to the will to live. The opposite of testosterone, in his framing, is not estrogen but prolactin, which makes people quiescent. "Testosterone makes effort feel good, estrogen makes emotions feel OK."

On fasting he separates mouse data, personal practice, and anecdotes. His former San Diego colleague Satchin Panda, author of *The Circadian Code*, published work in journals such as *Cell* showing that mice allowed to eat as much as they wanted, but only in a six- to ten-hour window aligned with light, stayed healthier and leaner than mice given the same calories spread across 24 hours. Ori Hofmekler had popularized a one-meal "warrior diet" earlier. In humans, fasting or low complex carbohydrate raises alertness via epinephrine. Foods rich in tryptophan — rice, grains, turkey, white meat — nudge toward sleepiness. The trap of one meal a day is that the meal becomes huge: a loaded gut shuts the system down. Huberman prefers a six- to eight-hour eating window. Low-carb is not automatically better; bulky fiber is miserable on a run, and carbohydrate carries water, which is why some people feel lighter and more explosive on a meat-heavy plate.

He almost regrets not opening with electrolytes. Neurons fire by rushing sodium inward; without enough extracellular sodium the nervous system does not run. Unless you have hypertension and provided you drink water, salt is useful, and it had been unfairly demonized. People who get shaky while fasting often think they need sugar; salt in water frequently fixes it. Huberman eats carbohydrates at night when he wants to be sleepy, including pasta, against West Coast fashion. The vegan-versus-carnivore fight, he thinks, has crowded out a large middle in which timing and personal experiments matter more than a tribe.

Asked what dreams mean, Huberman says part of the content is now concrete. Matt Wilson at MIT showed in rodents — later extended to non-human primates and humans — that hippocampal place cells replay the same sequence in sleep that they fired while navigating a maze. London taxi drivers, in the era before phones and GPS, were famous for mental maps of the city; analyses of their place-cell firing in sleep and waking fit the same picture. Replay is not mainly so you remember the route. It is so that, if there is a reason to remember — a crash at an intersection, a day stuck on one avenue while a spouse yells — spatial context gets bound to the emotional and hypothalamic systems. You never quite forget that time of day, that season, that event.

Early-night dreams, when slow-wave sleep dominates, tend toward general themes and locations, lighter on emotion, sometimes ending in a jolt awake. That same early window is when growth hormone is released. In the 1980s and 1990s, GHB — later illegal — became a party drug in part because it drove a massive growth-hormone release and a hypnotic state. REM dreams toward morning are a different species. Epinephrine in the brain is low, so you are paralyzed and cannot act the dream out, but the feeling is intense and loaded with theory of mind: you assign motives to other people. Huberman credits Simon Baron-Cohen — Sacha Baron Cohen's cousin, at Cambridge, not Oxford — with the framework. Deprive people of REM for too long and they become irritable and start attaching generally negative emotion to almost everything.

"The dreams that occur in REM sleep are when we divorce emotion from our prior experiences," Huberman says, and when we extract general rules. Susumu Tonegawa's work suggests hippocampal replay is an approximation of prior episodes; a lot of fear unlearning happens there. Wake suddenly from REM and you can get a surge of epinephrine and a racing heart — the fear arriving as you exit. Cannabis users, he notes, more often experience sleep paralysis, an invasion of REM atonia into waking: a terrifying second of being unable to move, then a jolt. Functionally he treats REM as self-induced trauma therapy. EMDR shuts down the amygdala with eye movements in the clinic; ketamine, a dissociative related to PCP, is used in some emergency settings to uncouple bodily sensation from a horrific event. PTSD, in this framing, is a failure of that uncoupling. The therapeutic aim is permanent: the person can recount the event without the same somatic terror.

REM itself was discovered in the 1950s at the University of Chicago: intense brain activity, high metabolism. Baron-Cohen's classic test for theory of mind has a child watch another child hide a toy and leave, then a second child enter; typical kids say the newcomer does not know where the toy is, while autistic kids often answer with the fact, "it's in the drawer." REM dreams are heavy with that mind-reading layer; slow-wave dreams almost lack it. During the Goggins event Fridman should not expect much REM unless he is extremely deprived, Huberman says, because high muscular output biases the system toward slow-wave recovery. What stops people in endurance, unlike a 500-pound deadlift, is usually a mental barrier. A Janelia Farm paper found that animals quit a running task when they are tricked into thinking their effort is futile, at a threshold of brainstem epinephrine; drop that level or give dopamine and they keep going. Humans, unlike mice, can borrow from other times — a person who has died, a future self — which is a reserve mice probably do not have.

Fridman connects dreaming to psychedelics through Lisa Feldman Barrett's intuition about dissociation and through Matthew Johnson's group at Johns Hopkins, which reached out after an earlier Huberman episode. Huberman is enthusiastic about that work. A former Stanford postdoc in the Hopkins orbit put octopuses on MDMA and found they wanted to spend more time with other octopuses, even cuddling — a *Current Biology* paper he mentions with a grin. Funding, once mostly private philanthropy, is starting to interest NIH. The cultural hangover is the 1960s: he jokes that you do not want to "Ken Kesey out of the game," after researchers who made themselves the experiment and lost institutional jobs.

Hallucinogenic states sit close to REM dreaming, a relationship Alan Hobson at Harvard Medical School mapped in books such as *Dream Drugstore* — among the first neuroscience books Huberman read in high school. Psychedelics can unveil plasticity: space-time relations loosen, emotional load can be reframed. Huberman's first objection is that people talk as if plasticity were the goal. Plasticity is a state in which you can direct change. The question is which circuits, toward what end. Opening the window with no destination is, in his view, a serious problem.

Chemists are already editing the molecules. A UC Davis paper in *Nature* a couple of months before the taping showed psychedelic compounds modified to drop the hallucinogenic component while keeping the plasticity piece. Some people will call that no fun; Huberman thinks it could matter for patients who would never take the classic drugs. MDMA is the odd one in the kit. Psilocybin and LSD drive heavy serotonin; MDMA drives dopamine *and* serotonin into the stratosphere at once — pursuit and contentment stacked, a combination he says has barely existed in human history. That is a live hypothesis for why it may help some depressions. The bigger conversation, if clinics are going to use these tools, is what the state is *for*: self-love, love of another person, forgetting hate. A physician friend put it in a line Huberman wishes he had coined: "A better living through chemistry still requires better living." Johns Hopkins' strongest results on smoking and drinking often showed up in the weeks of integration after the session. The drug loosens the brain; then you still have to do the work.

Huberman keeps returning to activities in which language is not enough: jiu-jitsu, surfing at dawn, dancing, even staring at a wall. Jiu-jitsu is nonverbal, and then people gather to talk about the nonverbal thing. The danger of the phone is not only online combat. Reading puts a script in your head, so the language system never shuts off. Wordless states, he argues, are as renewing as sleep and are a door into creative algorithms the brain will not run while it is composing something for someone else to understand. Sleep is one such period. Waking NSDR can be another. With or without drugs, the brain can enter modes that have no interest in being explained.

"Creativity is in the nonlinearities," he says, "but productivity is in the implementation of linearity." That is why formal training still matters: someone looks at the work and says it is not good enough, go do it again, or else ideas stay vapors. Johnson's group, Fridman notes, is even considering studies on creatives without a clinical condition. Huberman is skeptical that more psychedelics automatically help virtuosos. Jazz is nonlinear, but the players still had to learn the instrument. The anxiety of a highly linear person is letting go of the line; the anxiety of a highly creative person is being forced onto one. The scarce skill is toggling. Steve Jobs is his example of someone who talked about merging art and engineering. Psychedelics tend to shake the box; they are not a specific tool for going linear or nonlinear on command.

The plasticity mechanism Huberman thinks is best supported and least used in the real world is acetylcholine released from nucleus basalis in the forebrain — work associated with Mike Merzenich and his scientific offspring. Raise acetylcholine at the same time as a sensory, motor, or musical event, and neocortex reorganizes so that the event gets a permanent map. That is a gate on adult plasticity and even on single-trial learning. Eddie Chang, chair of neurosurgery at UCSF and a Merzenich trainee, is using the principle in studies of rapid language acquisition. Huberman's guess about why it never became standard in sport, music, or language training is cultural: scientists publish and move on; Merzenich talked about applications, then started a company, and the field backed away. Huberman thinks Merzenich was early, and that the timing is better now, including for kids with language deficits and dyslexia, not only for "optimization."

How to leverage it is the intersection of a physical practice and safer pharmacology. Smokers long ago learned that nicotine sharpens focus; lung cancer is the barrier. A Nobel-winning colleague at Columbia, unnamed, chews about six pieces of nicotine gum in a half-hour conversation, having switched from cigarettes. Huberman is not recommending Nicorette. He is saying directed nicotinic pharmacology, then machines plus drugs plus behavior, is where skill learning is headed. Until those pieces merge, he says, the field walks the same circle.

Fridman has downloaded the Joe Rogan video archive, extracted faces, and tracked blinks, throwing out glasses, smoke, baseball-cap shadows, and heavy eye makeup to keep the signal clean. He is hunting for cognitive-load markers that work in the wild. Huberman says blink rate rises with fatigue and falls when people are highly alert, and that strong data show each blink resets the perception of time, with a dopaminergic mechanism in the blink circuitry. Cross-person comparisons are mostly useless. Some people are hooded, some hardly blink, some are accused of being sociopaths when they may just run hot autonomic arousal. You need a baseline for that individual. The rarer frontier is two people at once. Neuroscience has mostly imaged isolated subjects in a scanner with a bite bar Huberman calls medieval. Two interacting brains can be treated as one system, signals highly correlated; he jokes that your blinks may trigger his. In a fight standoff, who blinks first is a real parameter of who can hold attention longer — not the only parameter, but a real one.

The hardest design choice on *Huberman Lab*, a show Fridman nudged him to start, was to stay on one topic for three or four episodes before switching. University teaching taught him that the gems are in the drill, not the pivot; the challenge is keeping it interesting, offering tools, and staying close to data. Listeners arrive with no prerequisites. Some want more nuclei and molecular pathways; others are already lost. He treats YouTube comments as office hours every third or fourth episode. Preparation is heavy: one recording a week, 10 to 15 hours of passive listening to university-level lectures (NIH has a public bank), 10 to 15 papers with a bias toward annual reviews, a whiteboard of intersections drawn while he lifts at home, then an outline he rehearses walking the dog. Five or six hours of that is active. Crystallizing the material hits his dopamine circuit hard. The only pain is knowing that a year later he would say it in half the words — and that at some point you still have to pull the trigger.

Fridman is considering a startup and thinking about Elon Musk, who has said his life is hell, that most people would not want it, and that he does it because something is probably wrong with him. Huberman has never met Musk; mutual friends from years ago speak highly of him. He thinks it is fair to say Musk accomplishes more before 9 a.m. than most people do, and that the load — scripted linear prediction across multiple companies, intense neurochemical output — would dissolve most people into a puddle of tears. Tiny issues such as coffee machines have been stripped away. If Neuralink staff could work at 10 percent higher capacity, or focus 5 percent better for 20 percent of the day, the time to goals would shrink, which might even lower the stress on Musk unless he started something else. Huberman is pushing 50; he thinks Musk is around 48. Sustaining that output for more than a decade is rare. Musk has said on Clubhouse that he worries his brain will someday fail under the puzzle load. Huberman's caution is the opposite move: taking someone in that regime and putting them where there is nothing to bite. His English bulldog Costello is happiest chewing and tugging a bone. "He is just not going to become a retriever."

Fridman's fear is more personal. He loves his current life, and for the first time he has caught real heat online — the child's question, why are these people making things up about me? A big company, if it works, means an army of people inventing stories even if most support you. Huberman does not soothe him. He says the answer may not be in a paper. It may arrive in the nonverbal, non-analytic 48 hours with Goggins, when you are forced to ask where your capacity is and whether you are even close. "I think we find the answers to those things in those non-verbal, non analytic states. It just comes to us." Fridman hopes it will be a fulfilling experience rather than one that leads to his demise. Huberman says that is exactly where the whole conversation converges.