Creatine, Sleep Trackers, and AI Doctors | Bryan Johnson Podcast

Kate shared Andrej Karpathy's two-month experiment comparing Oura, Whoop, Eight Sleep, and Apple. He found Oura and Whoop best. His key insight: sleep scores strongly predicted his work quality. Low scores meant lack of agency, courage, creativity; high scores enabled 14-hour flow states. The effect was cumulative—a few bad nights in a row were detrimental. Bryan framed this as intelligence augmentation: by measuring sleep, Karpathy was literally building his own intelligence. Bryan argued that in an era of AI, improving our own cognitive capacity through basics like sleep is the most leveraged action we can take. The protocol is simple: pick a tracker, wear it nightly, review data each morning, and notice correlations with your mental state. Over time, you'll develop body awareness and be able to adjust behaviors (bedtime, alcohol, evening light) to optimize scores.

Mike Min, a former ER doctor, explained that US ERs are overwhelmed—doctors see 30+ patients a day, leading to burnout and a system that defaults to the easiest path. He said if a loved one were in the ER, he'd input their complaints into an LLM and ask what should be happening, then ensure it happens. Kate demonstrated this with her mother's eye issue: she used ChatGPT to conduct a patient intake, asked about 100 questions, and got a list of recommended exams. The LLM correctly suggested corneal inflammation/scratch, which was later confirmed by an optometrist. Mike noted that ChatGPT already outperforms physicians on standardized tests, so for standard protocols it's reliable. This practice challenges the stigma against 'Googling' symptoms, reframing it as informed self-advocacy in a broken system.

Mike Min, Bryan's doctor, broke down the study's methodology. The researchers attempted a 7-day 'wash-in' period to account for water weight, but muscle creatine saturation actually takes 3-4 weeks at 5g/day, so the wash-in was too short. The subjects were untrained, and 12 weeks is a brief training window. More importantly, the study only measured lean mass, not strength, power, recovery, or cognitive function—the areas where creatine's benefits are most robust. Mike emphasized that the data for muscle mass gain was never strong; the real value is in longevity, brain health, and performance in well-trained athletes. He concluded that the study changes nothing for his recommendations.

Kate read Karpathy's findings: Oura and Whoop were top tier. He stated with 'absolute certainty' that his sleep scores correlated with work quality. Low scores meant lack of agency, courage, creativity; high scores enabled 14-hour flow states. The effect was cumulative—a single bad night was tolerable, but several in a row degraded performance. Bryan saw this as a perfect example of using measurement to augment intelligence. He argued that we are evolving into a new species and that improving our own intelligence through self-care is the most important thing we can do. Karpathy, an architect of superintelligence, demonstrated that optimizing sleep is a direct investment in cognitive capability.

Mike described his transition from ER medicine, where 80-90% of cases are preventable chronic disease, to longevity practice. He noted that ER doctors are 'getting absolutely murdered' and often burned out. When Bryan asked how to help a loved one in the ER, Mike said he'd input symptoms into an LLM, ask what should be happening, and ensure those steps occur. Kate shared how she used ChatGPT for her mother's eye issue: she had the LLM do a patient intake, asked 100 questions, and got a list of recommended exams. The LLM correctly suggested corneal inflammation/scratch, which was later confirmed. Mike cautioned that LLM quality depends on the user's ability to provide good information, but noted ChatGPT already outperforms physicians on standardized tests.

Bryan described a Kernel study where participants consumed no, low, medium, or high amounts of alcohol. At low and medium doses, behavioral tests showed no impairment, but brain scans revealed the brain was working harder to compensate. At high doses, compensation failed and both behavior and scans showed impairment. Bryan connected this to the casino observation: people consuming junk food, alcohol, and smoking appear to function normally, but their bodies are compensating until they can't. This underscores the value of granular measurement over subjective feeling. Mike added that once you become healthy, you become acutely aware of how bad unhealthy choices make you feel, creating a contrast with the general population who are adapted to chronic low-level impairment.