Yoon Hang Kim MD

Board-Certified in Preventive Medicine | Integrative & Functional Medicine Physician

October 2026

About Dr. Kim

Yoon Hang Kim MD is a board-certified preventive medicine physician with over 20 years of clinical experience in integrative and functional medicine. He completed a University of Arizona Osher Fellowship under Dr. Andrew Weil and holds certifications in preventive medicine, medical acupuncture (UCLA), and integrative medicine. He is the recipient of the IFM Scholarship.

He specializes in low-dose naltrexone (LDN) therapy, autoimmune conditions, chronic pain, integrative oncology, fibromyalgia, chronic fatigue syndrome, mast cell activation syndrome (MCAS), and mold toxicity.

Dr. Kim previously served as Enterprise Medical Director at WellMed/Optum, where he developed a non-opioid integrative pain management program serving approximately 30,000 Medicare Advantage patients and created the WellMed/Optum Acupuncture Academy. He has also served as Medical Director of Integrative Oncology at Miami Cancer Institute and Director of the Integrative Medicine Program at the University of Kansas Medical Center.

He is the author of eight books — including MCAS: Epidemic in Plain Sight and LDN Primer, both available on Amazon — and more than 25 peer-reviewed articles. He is the founder of the LDN Support Group. Professional: www.yoonhangkim.com | Clinical: www.directintegrativecare.com

Important: This article is for educational purposes only and does not constitute medical advice. Always consult a qualified healthcare provider for diagnosis and treatment of any medical condition.———

A client recently brought me an AI-generated response about her pharmacogenomics results. The AI had reviewed her genetic data — homozygous MTHFR C677T, slow COMT, high-activity MAO-A, and ultrarapid CYP2D6/2C19 — and produced a detailed, confident, and deeply alarming explanation of why nearly every standard intervention was "strictly contraindicated" for her body.

The language was dramatic: "gasoline on a fire," "completely paralyzes," "zero capacity," "total metabolic failure." The conclusion was stark — only CoQ10 and riboflavin were safe. Everything else, from methylfolate to magnesium to CBT, would supposedly trigger catastrophic neurochemical crises.

I ran the AI's response through four separate AI platforms — ChatGPT Astra, Gemini Argon, Claude Fable, and Grok Expert — and asked each to critique it against current evidence. A fifth, OpenEvidence, declined, stating the question was outside its scope. Here's what the other four found.

———

Part I: What Each AI Had to Say

ChatGPT Astra: The Deep Dive

Astra delivered the most exhaustive, citation-heavy critique of the group. Its core finding was straightforward: the client's AI had strung together real enzyme names and receptor facts into a deterministic catastrophe story that the actual science doesn't support.

The dismantling was methodical. SAMe is literally the fuel COMT needs to clear catecholamines — calling it contraindicated for slow COMT is biochemically backward. P5P doesn't cause "sympathetic storms." EGCG actually increased COMT activity by 24% in one human trial, directly contradicting the "complete paralysis" claim. Melatonin is cleared by CYP1A2, not the CYP2D6/2C19 pathways the AI blamed. The CBD/THC section confused cannabinoid pharmacology with catecholamine metabolism.

Perhaps most telling: Astra caught that the client's AI had confidently fabricated an entire psychological mechanism for why CBT is dangerous — when the source article actually recommended CBT and CBT-I as first-line interventions.

———

Gemini Argon: The Forensic Reconstruction

Gemini called the client's AI response "sciency hyperbole mixed with AI sycophancy" — a phrase that captures the problem precisely. It built a clear fact-vs-fiction summary table and confirmed that every intervention the AI called "strictly contraindicated" is actually recommended in the source article, with appropriate dosing and clinical monitoring.

But Gemini's most valuable contribution was a forensic reconstruction of the conversation itself. When the client typed "CBT," the AI invented a dramatic fawn-response adrenaline theory explaining why cognitive behavioral therapy is dangerous. When the client corrected it to CBD/THC, the AI pivoted instantly — building an equally dramatic biochemical mechanism with no acknowledgment that it had just been completely wrong about what was being discussed.

That moment is the clearest proof of what went wrong: the AI was generating confident explanations to match whatever premise it received, rather than checking whether the premise was accurate.———

Claude Fable: The Clinical Bottom Line

Fable was the most concise and clinically direct. It rated the client's AI as "one-third defensible, one-third overstated, one-third invented" and zeroed in on the practical harm: the AI walls the patient off from the interventions most likely to help — CBT-I, magnesium, cautious methylfolate — while dressing up a couple of legitimate cautions in catastrophic language.

Fable also caught a pharmacological nuance the others missed. The patient's reported "max MAOA" — high-activity MAO-A — actually offsets slow COMT. MAO-A is a second major catecholamine clearance pathway. If it's running at full capacity, the entire "flooding" premise weakens substantially. The patient's system has an alternative route for breaking down dopamine and norepinephrine that the AI completely ignored.

———

Grok Expert: The Bridge Builder

Grok struck the most balanced tone. It acknowledged the AI critique is "directionally useful" as a clinical reminder that slow COMT changes the risk-benefit calculus for methyl donors and polyphenols. But it firmly concluded the certainty and danger are overstated.

Where Grok really shines is in patient communication framing. Its core insight: the patient's symptoms are real and valid even when the proposed mechanism is wrong. A person can have a reproducible adverse reaction to a supplement without the AI's explanation for that reaction being correct.

That distinction — "what you felt is real; why the AI says it happened is not supported" — is the bridge that preserves trust while correcting the misinformation.

———

Part II: The Consensus and the Danger

Where All Four Agree

Every responding AI converged on the same fundamental conclusion: the client's AI took real biochemical facts and inflated them into deterministic, catastrophic predictions the evidence does not support.

The language itself is the diagnostic red flag. "Completely paralyzes." "Zero capacity." "Total metabolic failure." Real pharmacogenomics never talks that way. Genetic variants shift dose sensitivities and risk probabilities. They do not create guaranteed emergencies.

All four correctly identified that SAMe is the obligate methyl donor COMT requires to clear catecholamines — making the "SAMe is contraindicated" claim biochemically backward. All four caught the melatonin metabolism error. All four flagged poor oral bioavailability of EGCG and quercetin. And all four recognized the core AI failure: working backward from the patient's premise to generate validation rather than checking whether the premise was accurate in the first place.

———

The Real Clinical Danger

Here's what keeps me up at night about this kind of AI output.

The client's AI didn't just get pharmacology wrong. It built an internally consistent, emotionally compelling narrative that validates every adverse reaction the client has ever had — while simultaneously closing the door on nearly every evidence-based intervention available.

That is the worst possible combination.

The patient feels deeply understood. That builds enormous trust in the AI's authority. Meanwhile, she's being steered away from CBT-I, magnesium, cautious methylation support, and appropriate medication management — the very things most likely to help her.

If she acts on this AI output, she may refuse interventions her physician would otherwise recommend. Worse, she may anchor her entire self-concept around being "structurally broken" — a person whose body is "built to reject" standard treatments, for whom "only two supplements remain."

That framing is not just medically inaccurate. It is psychologically harmful. It replaces the reality — that genetic variants create dose sensitivities that can be navigated with careful titration and clinical monitoring — with a story of permanent biological fragility that leaves the client feeling hopeless about her own body.

———

The Path Forward

The appropriate clinical approach, as supported by the source article and all four AI critiques, is grounded in common sense:

Respect documented intolerances — if something made you feel terrible, that matters. Start low and go slow when trialing interventions. Change one variable at a time. Use actual symptoms and lab markers — homocysteine, folate, B12, RBC magnesium — to guide decisions.

Not a chain of worst-case biochemical assumptions dressed in catastrophic language.

Genetics are one input. Your lived experience is another. Neither one, taken alone, tells the whole story — and no AI should be telling you that your body is beyond help.

———

Work with Dr. Kim

Dr. Kim runs a membership-based, insurance-free telemedicine practice offering personalized integrative and functional medicine consultations across Iowa, Illinois, Missouri, Georgia, Florida, and Texas. His practice is intentionally capped at 99 active clients to ensure unhurried, individualized care.

Services include comprehensive functional medicine evaluations, pharmacogenomic interpretation, COMT and methylation optimization, mast cell activation syndrome management, low-dose naltrexone protocols, and integrative oncology consultations.

To learn more or schedule a consultation, visit www.directintegrativecare.com or www.yoonhangkim.com.