No Excuses

For the entire history of the species before about 2023, the limiting factor on improving your life was access to good information. The information either did not exist, or it existed somewhere you could not get to, or it existed in a form you could not understand without paying a specialist who had spent a decade learning to read it. If you wanted to know what your blood markers meant, you saw a doctor. If you wanted to know how to train your body, you hired a trainer. If you wanted to know what to eat, you asked a dietitian. Each of these gates cost real money and gave you a limited slice of expertise filtered through whatever the gatekeeper had been trained to say.

That gate is gone. There is now, sitting in your pocket, a system that will read your labs, design your training program, audit your diet, explain your medications, summarize the relevant research, write you a meal plan, give you a second opinion on your doctor’s first opinion, and hold your hand through the implementation, in plain English, for free, immediately, on demand, without a referral. It will do this for any topic you can name, at a level of competence that would have cost you tens of thousands of dollars in 2019. The cost of knowledge has gone to zero. Ignorance is no longer a valid excuse for anything.

You have to ask. That is the whole catch. You have to actually open the thing and type the question. Most people will not. Most people will keep paying the old-world gatekeepers and complaining that nobody helps them, while the most powerful intellectual tool in human history sits unused in their browser tab. Do not be them.

A warning, before you get comfortable. The AI you have access to today isn’t the AI you’ll have access to in five years.

Every digital platform in human history has followed the same arc. It gets very good at serving users in order to grow, it gets popular, and then, the moment the user base is captive, it pivots to monetizing the users instead of serving them. Cory Doctorow called this enshittification. Search engines, social networks, marketplaces, dating apps — every one of them ran the script. Google search in 2008 was a miracle. Google search in 2024 is a maze of ads, “sponsored results,” and SEO sludge that returns less useful information than a well-made library card catalog from 1985. The technology did not get worse. The incentives did.

AI is going to run the same arc. Right now we are in the good phase — the providers are competing for users, training is improving fast, the answers are surprisingly clean, and there is no obvious advertising layer between you and the model. Within a few years that will change. Recommendations will get injected. “Sponsored insights” will appear. The model will quietly steer you toward products its training partners have an interest in. The ad vector will become the dominant economic model, exactly as it did everywhere else, and the same captured information environment that already eats search and social will eat AI. You will not notice it happening, because the change will be gradual and the answers will still feel useful, just a little off, just a little aligned with something other than your interests.

This is also why I do not lie awake about superintelligence. The actual near-term risk from AI is not that it becomes a god and decides to kill us. The actual near-term risk is that it becomes a captured advertising channel that decides what we believe, and we love it for telling us what we already wanted to hear, and the same forces that ate every other media channel eat this one. The threat is corporate, not robotic. It is also boring, which is why almost no one talks about it.

So use it now. Use it hard. Get what you need out of it while the window is open and the answers are still mostly aligned with your actual interests. The next few years are a free pass on a tool that will not stay free.

Most people use AI badly. They open the chat, type a one-line question — “what’s the best diet” or “how do I lose weight” — and they get a one-line answer, which is necessarily generic, hedged, and useless. Then they conclude that AI is overrated.

The unlock is multi-paragraph prompts. You give the AI your situation in detail. Your age, your numbers, your symptoms, your history, what you have already tried, what your constraints are, what your goals are, what answer you do not want (the generic one). You tell it who you are and what you actually need. The quality of the answer goes up by an order of magnitude.

A single-line prompt:

What should I eat for autoimmune?

A real prompt:

I am a forty-five-year-old woman with relapsing-remitting multiple sclerosis, diagnosed eight years ago, currently on no DMTs by choice. My fasting insulin is 12, HbA1c is 5.6, hs-CRP is 3.2. I have been carnivore for nine months and feel dramatically better but still get episodes of fatigue around my cycle. I take no medications other than two supplements (magnesium glycinate, vitamin D3 5000 IU). My goals are to keep reducing systemic inflammation and to extend remission as long as possible. I want a ranked list of dietary, sleep, exercise, and supplement interventions I have not yet tried, with the strongest evidence first, with explicit notes on which interventions are backed by clinical data versus mechanistic reasoning versus anecdote. Do not recommend anything without naming the evidence tier.

The first prompt gets you a Wikipedia paragraph. The second prompt gets you a printable plan that a competent functional medicine doctor would charge four figures to produce. The information is the same. The way you asked for it is everything.

Then there is the second move, which is even more important: ask the AI to refute itself. Once it has given you an answer, paste the answer back and ask it to take the opposite position. Ask it for the strongest argument against its own recommendation. Ask it to identify the ways it might be wrong, the evidence it might be misweighting, the assumptions it slipped in without naming. The model will, in most cases, do this honestly, and the contrast between its own confident answer and its own honest critique is where the actual learning lives. You are using a tool that is, uniquely in human history, willing to argue against itself on demand. Use it.

Using AI well is itself a skill, and the skill correlates with the same general intellectual capacities that everything else correlates with. People who are sharp, curious, and willing to read carefully get a lot out of AI. People who are not, do not. This is uncomfortable to say in a culture that wants every tool to be perfectly equalizing, but it is the truth, and ignoring it makes the gap worse, not smaller.

The good news is that the skill is improvable. You learn to prompt by prompting. You learn to spot bad answers by getting bad answers and noticing they are bad. You learn to ask the right follow-ups by trying the wrong ones first. After a few months of regular use, your prompts get better, your skepticism gets sharper, and the model starts feeling more like a competent collaborator and less like a smart-sounding parrot. The skill compounds, the same way every other skill compounds, and the people who start now will be much better at this in two years than the people who start in two years.

The use case I care about most, and the reason this essay sits on a website about MS and metabolism, is that AI is the first time in history a regular person can build something approaching elite understanding of their own body without paying a specialist for the privilege.

You can paste your full lab panel into the model and get a reading of every marker, in context, with the bad-but-still-in-the-normal-range traps called out and the optimal-range targets explained. You can describe your symptoms over the last six months and get a differential diagnosis list ranked by likelihood, with the specific follow-up questions to ask your doctor next. You can submit your training log and get programming critiques. You can ask for the strongest evidence-based interventions for your exact condition and get them, ranked, sourced, with caveats. You can ask “what does my insurance not want my doctor to tell me about my condition” and get a pointed, honest answer.

The leverage is enormous. The only people who do not get the leverage are the people who do not ask.

The end state of AI use, for personal health, isn’t lifelong dependence on the model. It is a phase. You use the AI heavily for six to twelve months, you build genuine understanding of your own body, your own numbers, your own protocols, and at some point the model starts telling you things you already knew. That is the signal. The leverage was the on-ramp. The destination is competence.

Then you teach the people around you, in person, with patience. You hold the next person’s hand through the lab they cannot read. You build the household around the habits the AI helped you find. The knowledge stops needing the machine. The habits do the work.

This matters because the window will eventually close, and when it does, the people who used the open window to actually become competent will be fine, and the people who outsourced their thinking to the model and never internalized the underlying material will be at the mercy of a captured tool. Use the bridge while it is there. Then walk off the bridge.

JUST FUCKING ASK

There is, sitting on your phone, the most powerful research assistant ever built, and it’s willing to work on any problem in your life for free, right now, and most people will never type the question.

Don’t be most people. Open the thing. Paste your numbers. Describe your situation. Ask the question you have been telling yourself you would ask “someday” when you got around to seeing a specialist. Ask it about your sleep, your training, your diet, your hormones, your medications, your kid’s symptoms, your spouse’s diagnosis, your legal situation, your taxes, your retirement plan, your mortality. Ask it the questions you are embarrassed to ask a person. Ask it the questions you would have to pay a person five hundred dollars to answer. Ask it three times in three different ways. Ask it to refute its own answer. Ask it to write you a one-page protocol you can hand to your spouse.

Then put the phone down and execute.

Just fucking ask. Then execute. Knowledge has been the bottleneck for the entire history of our species. It is not the bottleneck anymore. The new bottleneck is whether you are willing to do the work the answer tells you to do — and that part, the machine cannot solve for you.