ChatGPT Health is a marketplace, guess who is the product?
consciousdigital.org
consciousdigital.org
I want regulators to keep an eye on this and make smart laws. I don't want it to go away, as its value is massive in my life.
(One example, if you are curious: I've been doing rehab for a back injury for about 10 years. I worked with a certified trainer/rehab professional for many years and built a program to keep me as pain-free as possible. I rebuilt the entire thing with ChatGPT/Gemini about 6 weeks ago, and I've had less pain than at any other point in my life. I spent at least 12 hours working with AI to test and research every exercise, and I've got some knowledge to help guide me, but I was amazed by how far it has come in 12 months. I ran the results by a trainer to double-check it was well thought out.)
How is it we have come to a place in society where we second-guess everything we type? But perhaps also where we go (with our "tracking devices" in our pockets…).
I mean, obviously the internet is what changed everything. But it is like you have a megaphone strapped to your face whenever you connect to a site and make a comment.
Maybe this is not a good thing.
How is this manifesting in reality?
And the great thing about it is that you already signed all your rights away for them to do this exact thing, when we could have had an open world with open models run locally instead where you got to keep your private health information private.
If the goal is providing subsidies (i.e. wealth transfers), then insurance is not the way to do it. That is the government’s role.
And it naturally means the people with highest premiums are the least likely to be able to afford it (the elderly, the disabled, those with chronic conditions that make them less likely to maintain high earning jobs steadily, etc)
"If I focused on my health, ate clean and exercised daily, why should I also be subsidizing Billy "video-games-are-my-exercise" fatass's chronic health conditions?"
This is why there is a hyperfixation on shifting blame away from (failing) individuals. The logic breaks when Billy has to admit he just hates exercising.
And yes, before you comment, I know "maybe Billy has (condition outside all control) so it's not on him". Please, see what I just said in the previous statement.
I don't think we should play arbiter for who has and hasn't lived a healthy enough life to still believe they should get healthcare?
The battle along the spectrum of privatizing gains (lower healthcare premiums for a healthy lifestyle - high premiums for unhealthy lifestyle) vs socializing losses (paying $20/mo to get $1200/mo of care - paying $1200/mo for $0/mo of care) is constant and boundless in either direction.
On end, it's "national insurance", functionally equivalent to fully-tax-funded healthcare like the NHS or the German system with several providers competing but regulated to near identical results, but moreso as the UK and Germany also has private care; on the other, it's the absence of insurance.
Those might all seem wildly disconnected, but they all have systems of unfair allocation to compensate for unequal outcomes.
Generally national healthcare programs are entirely dependent on young healthy people paying into the system despite rarely needing it, and then hopefully enough dieing quick deaths or having multiple children to cover their costs. These rebalancing systems are artificial and humans are generally terrible at managing them.
> why should I also be subsidizing Billy "video-games-are-my-exercise" fatass's chronic health conditions?"
Nobody is asking you to: enrolling in insurance is a choice in the USA.
Also, replace "chronic health conditions" with "unavoidable inherited genetic risk factors". We don't want Billy to be screwed for life just because he was born to a suboptimal combination of parents.
The choice right now is all or nothing. There is no choice for healthy people to only share a plan with other healthy people. If there was, everyone else, especially the least healthy, would be totally screwed.
1. In the US, enrolling in health insurance is currently a choice (I'm can't tell whether your hypothesis about healthy people is agreeing or disagreeing with this point).
2. We don't want Billy to be screwed for life just because he was born to a suboptimal combination of parents.
Would you mind clarifying the connection here, please?
2. >This is why there is a hyperfixation on shifting blame away from (failing) individuals. The logic breaks when Billy has to admit he just hates exercising.
I never said they shouldn't. It's just a topic totally unrelated to whether or not people are forced to buy insurance, and I'm not interested in discussing that new, unrelated topic.
> Billy has to admit he just hates exercising.
This logic breaks down when Billy actually exercises and eats just fine, but was nonetheless born to a suboptimal combination of parents.
The system can accommodate (and frankly is the ideal many people strive for) some health nut getting long drawn out cancer battle at 41. Its rare enough to be noise in the giant money payout pool.
Obesity and it's litany of health effects are not rare, and next to age, are a dominating signal drawing money from the pool.
But such a system comes at other costs that most people intuitively feel infringes on core values they have.
Edit to add: this system would actually have some great advantages over an “existing conditions” tax in that now you pay low rates until you have diabetes, all during the time you are leading the unhealthy lifestyle. But once you have it you are not rewarded for starting to exercise and eat healthy and get it under control. In the hypothetical scenario above, you’d be punished economically during the period you were building bad habits and you would be able to restore sane costs after course correction
This is difficult to convey to certain brands of self-styled libertarians.
Which we have anyway. We might as well get some benefits from it.
Then why are you not asking your insurer why they cover a lot less preventative health or other options. For example, Kaiser flat out refuses to prescribe GLP-1s for weight loss, others insurers are the same with gym subsidies or not covering nutritionists.
But they'll happily pay for your gastric bypass.
That's true for predictable costs, but not true for unpredictable ones - which is the point of most insurance (housing, car, etc). The point and use of insurance is to move risk to entities that can bear it.
Utility is non-linear with money, and so you easily have situations where spending X times more on something "costs" you more than X times if measured in how useful the money is to you.
Typically, as you have more money, each further dollar doesn't provide as much benefit as the last (sometimes things are lumpy, the difference between "not quite enough to pay rent" and "just enough to pay rent" is huge, but broadly this is true). Going from $1000 to $10000 is more impactful than $1001000 to $1010000.
That means that moving the other way, each additional dollar spent has a greater personal cost to you.
Therefore, sharing unlikely but high expenses can mean that your expected cost is the same (if there's no profit/middleman) or a bit higher, but your expected personal cost is lower.
Even the age rating factor is capped at 3, so there are also massive wealth transfers from young to old.
Mathematically, health insurance premiums in the US are more tax than insurance premium.
AIUI the idea is to predict the "correct" price for an individual premium, Y, which is restricted to being the result of some insurance model function, f(), that is itself restricted to the domain of age (A), location (L), and smoking status (S):
Y = f(A, L, S)
But why is location useful for insurers to price premiums? I assume because healthcare has different costs in different locations, and different utilization rates: People living in coal mining towns or near oil refineries may be expected to use more healthcare and thus cost more to insure.
Thus, you can imagine insurers building a price map (like a heat map) overlay for the state/country, plotting each applicant on it, and checking the "color" there as part of their model function. So they are effectively embedding out-of-band information (prices and utilization rates for locations) into the model function f() itself.
What "AI", or large-parameter Deep Neural Networks, fundamentally change here is:
- They can approximate any effectively computable function, the same class of functions that a programmer can write and execute on a computer[0].
- They can be scaled up to an arbitrarily large number of parameters, i.e. an arbitrarily precise approximation limited only by the number of parameters and the amount and quality of model training data.
- Absolutely critically: They are programmed implicitly, through brute-force training on input-output pairs, rather than explicitly, with some programmer writing an explicit series of instructions to compute the function.
Large insurers previously had sophisticated models for f() hand-built by math whizzes, but they were limited in the amount of out-of-band information they could encode into that function by the limited cognitive and programmatic capacity of a human team.
But now with DNNs you can scalably encode unlimited out-of-band information into the function f(), while also totally obscuring how you're computing that location-based price adjustment.
The result, in extremis, is that f() is not some fancy algorithm cooked up by a few dozen math whizzes. Instead f() becomes a fancy database, allowing the tuple (A, L, S) to act merely as an index to an individualized premium Y, which defeats the entire purpose of restricting the domain of their model function.
[0]: https://en.wikipedia.org/wiki/Universal_approximation_theore...
Given the magnificent ability of DNN models to serve as obfuscatory black boxes and the general techno-ignorance of legislators and regulators, I suspect that "AI laundering" your violations actually a very effective way to juke all sorts of laws and regulations.
But both of us are just speculating. If you have insider industry knowledge or can point to regulatory guidance and/or enforcement actions in this area that corrects or confirms my understanding, I would love to read about it.
That has not been my impression as an outside observer.
https://www.healthcare.gov/coverage/pre-existing-conditions/
https://www.healthcare.gov/how-plans-set-your-premiums/
Health insurance premiums in the US are more tax than insurance. They also have low single digit profit margins with less than desirable shareholder returns (many are non profit in the first place), so they don’t have much room to lower premiums without also reducing healthcare expenses.
The insurance business in general is very competitive and not very profitable, so an insurer that tries to collect outsized premiums will usually suffer a loss of business.
Knowing several Americans, and how much they pay for health insurance (and are still required to pay for some things "out of pocket"), this is incredible to me. And that's before you even get to the process of making a claim.
Such a broken system.
When that didn't happen, the story changed to that number being how much more premiums would have risen.
Insurance premiums have only gone up as far as I can remember, though there's a ton of variables at play here. Inflation is an obvious one, plus continual introduction of more and more costly treatments- biologic injections, cancer therapies and so forth. The unfortunate increase in obesity rates in my lifetime (along with all the health complications) has been a significant contributor as well.
It all adds up.
That was eliminated by a Republican bill, the Tax Cuts and Jobs Act of 2017.
Most people get insurance through their employer, and most employer plans (at medium to large companies) are self funded by the company and merely administered by insurance companies.
That means the healthy participants had no effect on those plans whatsoever. Even at peak, the individual mandate had only cut the number of uninsured by half, and the effect on rates was negligible.
An interesting thing about rising health costs is that it has happened at roughly similar rates in most first world countries for the last 50+ years.
For example in 1990 the UK, FR, and US were paying 2.0, 2.2, and 2.6 times their 1980 costs per capita. By 2000 that was 4.1, 4.1, and 4.2. By 2018 (the last year I had data for when I calculated this a few years ago) it was 10.6, 7.5, and 10.2.
Here's the 2000 to 2018 increase for those and some others: DE, FR, CA, IT, JP, UK, US were 2.1, 1.8, 2.0, 1.7, 2.6, 2.6, 2.3.
When politicians in the US talk about rising health care costs they usually put the blame on recent policies from opposing politicians. That so many first world countries with so many different health care systems all have seen similar increases for the last 50+ years suggests that it is due to something they all have in common and that government policy doesn't affect it much.
Prices were always going to increase.
baggage fees,
or lack of WiFi,
on, say, Spirit Airlines.
You looked at the list of insurers/jobs offering health insurance carriers,
selected the cheapest, or second-cheapest option,
and you’re surprised they’re harder to work with?
How?
Even in tech spaces with money to throw around, that just means that maybe your partner and dependent's premiums will be covered/negligible, or that your deductible will be low.
You're still going to be fucked by their policies, though.
Pay me enough in cash to secure my own wellness with whatever organizations I choose, or: next, please.
Absolutely not. They inflate prices by 200% and then give you 20% "savings" back. The whole idea of a health insurance company as publicly traded corporation is totally insane. They are designed to extract maximum profit from wherever they can get. The is no incentive to save money for patients. Any savings go to shareholders.
If your claims were true, then the publicly traded businesses would have no customers.
Yeah. And they all ultimately have to be paid by my health premiums.
When I visit my parents, and they have the TV on, and I see 3 90-second advertisments for prescription drugs every single commercial break, I remind myself that, no matter what, we have to keep funding these commercials. Whatever the US decides to do for healthcare, I guarantee these commercials will continue to be paid for.
I also guarantee that pharmacy company executives, and insurance company executives will continue to make millions of dollars a year. We have to keep paying them as well.
The huge bureaucracy of insurance workers who decide what is and isn't approved, that all will have to be paid for as well.
I used to work for a company that did background checks on doctors, we had different customers in every state, every state had their own companies and their own system for maintaining and verifying doctors licenses. These different companies in each state have to be paid. I made good money as a programmer doing background checks for these various companies and my paycheck also ultimately came from your medical premiums.
I think we need to stop and appreciate the patriotic duty we all have to pay high premiums and medical costs, because every time we do we're propping up a huge portion of US workers. -- Just kidding. The truth is a lot of companies need to go out of business before things get cheaper.
You would not be privy to their internal processes, and thusfar not be able to prove wrong doing. You would just have to hope for a new Snowden and that the found wrongdoings would actually be punished this time.
Also, in some cases: they absolutely do. Try to get hired in Palantir and see how much they know about your browsing history. Anything related to national security or requiring clearances has you investigated.
For instance back when I was interviewing at startups and other companies where I was going to be a strategic hire, I would casually mention how much I enjoyed spending time on my hobbies and with my family on the weekend so companies wouldn’t even extend an offer if they wanted someone “passionate” who would work 60 hours a week and be on call.
But is it really so hard to imagine a world where your individual choice to "opt-out" or work for companies that don't use that info is a massive detriment to your individual life? It doesn't have to be every single company doing it for you to have no _practical_ choice about it (if you want to make market rate for your services.)
But the country and the people who could most benefit from it are more concerned with whatever fake outrage Fox News comes up with an anti woke something or the other.
So yeah, if this is the country America wants, I’m over it. I’ve done my bid.
While other people talk about leaving the country, we are seriously doing research and we are going to spend a month and a half outside of the US this year and I’ve already looked at residency requirements in a couple of countries after retirement including the one we are going to in a month and a half.
I think GP is suggesting that you're supposed to do something akin to what Ben Kenobi did while aboard the Death Star, not what he did beforehand.
This, in no way, represents my own feelings or opinion on this matter. I'm just trying to aid the conversation.
I basically did what they plan on doing. I fcked off because my country was already too far gone. But I always always make sure I will never talk positively or be in denial about the state it’s in. America isn’t there (yet). What made me snarky was the mistaken hypocrisy.
Calibrating your threat model against this attack is unlikely to give you any alpha in 2026. Hiring at tech companies and government is much less deliberate than your mental model supposes.
The current extent of background checks is an API call to Checkr. This is simply to control hiring costs.
As a heuristic, speculated information to build a threat model is unlikely to yield a helpful framework.
How could you possibly know if they use other undisclosed methods as part of the recruitment? You are assuming Palatir would behave ethically. Palantir, the company that will never win awards based on ethics
Notwithstanding the fact that tech companies hire dogshit employees all the time and the vast majority of employees of any company of size 1000+ are average at best, Palantir happens to be rating so high on the scale of evil that I'd pop champagne if it got nuked tomorrow.
If any company would do it, it would be Palantir.
Also, I’m not saying they have the best talent, just that they want the best talent.
It is impossible to prove a negative, but having strongly held beliefs without evidence is an antipattern.
Do you have any proof they don’t have a goose randomly deciding to hire you?
The lack of proof gives no credence to it actually happening
Under this conspiracy theory they'd have to be available for sale somewhere, right? Yet no journalist has ever picked up the story? Nobody has ever come out and whistleblown that their company was buying Google searches and denying applicants for searching for naughty words?
It is well documented in many many places, people just don't care.
Google can claim that it doesn’t sell your data, but if you think that the data about your searches isn't being sold, here is just a small selection of real sources.
https://www.iccl.ie/wp-content/uploads/2022/05/Mass-data-bre...
And it isn't paranoia, consumer surveillance is a very real problem, and one of the few paths to profitability for OpenAI.
https://techpolicy.sanford.duke.edu/data-brokers-and-the-sal...
https://stratcomcoe.org/cuploads/pfiles/data_brokers_and_sec...
https://www.ftc.gov/system/files/ftc_gov/pdf/26AmendedCompla...
https://epic.org/a-health-privacy-check-up-how-unfair-modern...
Citation needed for a claim of this magnitude.
> It is well documented in many many places, people just don't care.
Yes, please share documentation of companies buying search data and rejecting candidates for it.
Like most conspiracy theories, there are a lot of statements about this happening and being documented but the documentation never arrives.
https://www.upturn.org/work/comments-to-the-cfpb-on-data-bro...
> Most employers we examined used an ATS capable of integrating with a range of background screening vendors, including those providing social media screens, criminal background checks, credit checks, drug and health screenings, and I-9 and E-Verify.29 As applicants, however, we had no way of knowing which, if any, background check systems were used to evaluate our applications. Employers provided no meaningful feedback or explanation when an offer of work was not extended. Thus, a job candidate subjected to a background check may have no opportunity to contest the data or conclusions derived therefrom.30
If you are going to ignore a decade of research etc... I can't prove it to you.
> The agency found that data brokers routinely sidestep the FCRA by claiming they aren't subject to its requirements – even while selling the very types of sensitive personal and financial information Congress intended the law to protect.
https://www.consumerfinance.gov/about-us/newsroom/cfpb-propo...
> Data brokers obtain information from a variety of sources, including retailers, websites and apps, newspaper and magazine publishers, and financial service providers, as well as cookies and similar technologies that gather information about consumers’ online activities. Other information is publicly available, such as criminal and civil record information maintained by federal, state, and local courts and governments, and information available on the internet, including information posted by consumers on social media.
> Data brokers analyze and package consumers’ information into reports used by creditors, insurers, landlords, employers, and others to make decisions about consumers
https://files.consumerfinance.gov/f/documents/cfpb_fcra-nprm...
And that CFPB proposal was withdrawn:
https://www.consumerfinancialserviceslawmonitor.com/2025/05/...
Note screen shots of paywalled white papers from large HR orgs:
https://directorylogos.mediabrains.com/clientimages/f82ca2e3...
Image from here:
https://vendordirectory.shrm.org/company/839063/whitepapers/...
But I am betting you come back with another ad hominem, so I will stay in the real world while you ignore it, enjoy having the last word.
Every link you provided is for tangential things. They're bad, yes, but they're not google searches. Provide a link where some individual says "Yes, I know what so-and-so searched for last wednesday."
Can you answer this question without walls of unrelated text, ad hominem attacks (saying I’m in a cult), or link bombing links that don’t answer the question?
It’s a simple question. You keep insisting there’s an answer and trying to ad hominem me for not knowing it, but you consistently cannot show it.
Anyone who has worked in hiring for any big company knows how much goes into ensuring hiring processes don't accidentally touch anything that could be construed as illegal discrimination. Employees are trained, policies and procedures are documented, and anyone who even accidentally says or does anything that comes too close to possibly running afoul of hiring laws will find themselves involved with HR.
The idea that these same companies also have a group of people buying private search information or ChatGPT conversations for individual applicants from somewhere (which nobody can link to) and then secretly making hiring decisions based on what they find is silly.
The arguments come with the usual array of conspiracy theory defenses, like the "How can you prove it's not happening" or the claims that it's well documented that it's happening but nobody can link to that documentation.
For instance, if your job is to be on your feet all day and you can barely stand, then that job is not for you. I have never met employers that are so flush in opportunities of candidates that they just randomly choose to exclude certain people.
And if it's insurance, there's a group rate. The difference only variable is what the employee chooses out of your selected plans (why make a plan available if you don't want people to pick that one?) and family size. It's illegal to discriminate of family size and that does add up to 10k extra on the employer side. But there are downsides to hiring young single people, so things may balance out.
If we circumvent those privacy laws, through user licenses, or new technology - we are removing the protections of normal citizens. Therefore, the bad behavior which we already decided as a society to ban can now be perpetrated again, with perhaps a fresh new word for it to dodge said old laws.
If I understand your comment, you are essentially wondering why those old laws existed in the first place. I would suggest racism or other systemic issues, and differences in insurance premiums, are more than enough to justify the existence of privacy laws. Take a normal office job as an example over a manual labor intensive job. No reason at all that health conditions should impact that. The idea of not being hired because I have a young child, or a health condition, that would raise the group rate from the insurer passing the cost to my employer (which would be in their best interest to do) is a terrible thought. And it happened before, and we banned that practice (or did our best to do so).
All this to say, I believe HIPAA helps people, and if ChatGPT is being used to partially or fully facilitate medical decision making, they should be bound under strict laws preventing the release of that data regardless of their existing user agreements.
It’s not just medical but a broad carve out called “bona fide occupational qualifications”. If there’s a good reason for it, hiring antidiscrimination laws allow exceptions.
So less, the job requires you to stand all day, and more, once a week or so they ask you make a binder of materials, and the hole puncher they want you to use dislocates your hands (true story). Or, it's a desk job, but you can't get from your desk to the bathroom in your wheelchair unless they widen the aisles between desks (hypothetical).
Read your policy!
Insurers derive rates for each employer from each employer's costs where laws allow this. And many employers self fund medical insurance.
GDPR Request. Ah wait, regulation bad.
The continued secrecy of the conspiracy would then depend on every person involved in orchestrating this privacy violation and illegal hiring scheme keeping it secret forever. Nobody ever leaking it to the press, no disgruntled employees e-mailing their congress people, no concerned citizens slipping a screenshot to journalists. Both during and after their employment with the company.
To even make this profitable at all, the data would have to be secretly sold to a lot of companies for this use, and also continuously updated to be relevant. Giant databases of your secret ChatGPT queries being sold continuously in volume, with all employees at both the sellers, the buyers, and the users of this information all keeping it perfectly quiet, never leaking anything.
You just get an automated denial from the ATS that's based on the output from AI inference engine.
There is a vast gap between what is not legal and what is actually actionable in a court of law, which is well known to a large power nexus.
What are the chances those bills last long in the current administration and supreme court?
The US has been pretty much a free-for-all for surveillance and abusing all sorts of information, even when illegal to do so. On the rare occasions that they get caught, the penalty is almost always a handslap, and they know it.
If the person can use AI to lead a noticeably better life, something that may have been impossible previously due to economic circumstance, then the first order benefits outweigh the second order drawbacks.
I’m not disputing what you’re saying, I just think that treating it like a zero sum game every time the conversation comes up is showing an immense amount of privilege.
You, me, the parent commenter; we’re all dying, we don’t have time to optimise for the best outcome.
I know that neither health insurers nor any government agency nor anybody else have even 0,0000000000000001% as much interest in my health, well being and survival as I do.
When it is the matter of my health and my life, I care as much about what an insurer or employer thinks as I would care about what the Ayatollah of Iran thinks. Or what you think. Ie: Those opinions are without any value at all.
So if insurers can cut you off based on your ChatGPT queries or test results then you may find yourself in serious debt, homeless, without medical care, etc
Bankruptcy is of course much preferable to not having your health. Even having to argue this is bizarre. We are not human batteries in the Matrix. Our purpose is not to please institutions or destroy our own lives for fear of hypothetical situations.
Don't you understand that you only have one life and one body. That's it. You have 70 or 80 years with one body and one mind. That is the only thing which matters.
We have seen, again and again, business encroach on our quality of life and it does warrant skepticism and alertness about their motivations.
But we can have that? If you have powerful enough hardware you can do it, right now. At very least until the anti-AI people get their way and either make the models' creators liable for what the models say or get rid of the "training is fair use" thing everyone depends on, in which case, sure, you'll have to kiss legal open-weight models goodbye.
This is probably a field that MistralAI could use privacy and GDPR as leverage to build LLMs around that.
Then you start "digging deeper" on a specific sub-topic, and this is where the risk of an incorrect response grows. But it is easy to continue with the assumption the text you are getting is accurate.
This has happened so many times with the computing/programming related topics i usually prompt about, there is no way I would trust a response from an LLM on health related issues I am not already very familiar with.
Given that the LLM will give incorrect information (after lulling people with a false sense of it being accurate), who is going to be responsible for the person that makes themselves worse off by doing self diagnosis, even with a privacy focused service?
Like you, I fact-check it (well, search the internet to see if others validate the claims/points) but I don't do so with every response.
LLMs are yet another powerful tool under our belt, you know it’s hallucinating so be careful. That said, even asking specialized info about this or that medical topic can be a great thing for patients. That’s why I believe it’s a good thing to have specialized LLMs that can tailor responses on individual health situations.
The problem is the framework and the implementation end goal. IMO state owned health data is a goldmine for any social welfare system and now with AI they can make use of it in novel ways.
https://arstechnica.com/health/2025/08/after-using-chatgpt-m...
So what's your argument?
But framing it as a “myth [of] perfect attention and treatment” sounds a bit like delegitimizing the entire healthcare industry in a way that makes me raise my eyebrow.
Positive outcomes per patient is probably also higher, due to research and technology advances. So many lives saved that would have been written off just a decade or two ago (e.g. spina bifida).
But I agree with you that there’s a hypothetical universe where seeking healthcare as an American doesn’t suck, I just don’t know if “productive” is the right word to describe it.
It doesn't delegitimize the whole industry. It points out real problems. A lot of patients are not given enough attention and don't get the correct treatment because the doctors didn't listen but rushed through things.
The “myth [of] perfect attention and treatment” is an easy strawman for grifters and conmen to take advantage of: see RFK Jr.
I have no idea what other professions you’re talking about. Doctors are the only professionals where it’s common for multi million dollar judgements to be awarded against individuals. In may cases, judgements larger than their malpractice insurance limits.
Take a doctor working alone overnight in the ER. They are responsible for every single thing that happens. One of the 4 NPs that they are supposed to have time to supervise while they are stuck sedating a kid for ortho to work on makes a mistake—the doctor is the one that’s getting sued. A nurse misinterprets an order and gives too much of something, doctor is getting sued. Doesn’t matter if it’s their fault or not. Literally ever single one of the dozens of patients that comes in with a runny nose or a tummy ache, or a headache is their responsibility and could cost them their house. And there are far too many patients for them to actually supervise fully. They have to trust and delegate, but in practice they are still 100% on the hook for mistakes. For accepting this responsibility they might get $10 per NP patient that they supervise.
Healthcare professionals also occasionally face criminal prosecution for mistakes at a level that wouldn’t even be me a career in other professions.
> Lawsuits against medical professionals are difficult in many cases impossible for the average person to win
Malpractice attorneys operate on contingency, so they’re more accessible to the average person than most kinds of attorneys. It’s one of the many reasons healthcare is so expensive in the US.
It’s harder for a doctor to get fired for saying showing up late to work than it is for a cook at McDonald’s I guess, but compared to other professionals? I’ve seen software engineers regularly skip through companies leaving disasters in their wake for their entire careers. MBAs regularly destroy companies, lawyers and finance bros get away with murder, and police officers literally get away with murder.
The only profession that faces anywhere near the accountability that doctors do that I can think of might be civil engineers.
If you are treated by a human being and it goes sideways, you could sue them and/or the hospital. Now, granted, you may not always win, it may take some time, but there is some chance.
If you are "treated" by an LLM and it goes sideways, good luck trying to sue OpenAI or whoever is running the model. It's not a coincidence that LLM providers are trying to put disclaimers and/or claims in their ToS that LLM advice is not necessarily good.
Same goes for privacy. Doctors and hospital are regulated in a way that you have a reasonable, often very strong, expectation of privacy. Consider doctor-patient confidentiality, for example. This doesn't mean that there is no leak, but you can hold someone accountable. If you send your medical data to ChatGPT and there is a leak, are you going to sue OpenAI?
The answer in both cases is, yes, you should probably be able to sue an LLM provider. But because LLM providers have a lot of money (way more than any hospital!), are usually global (jurisdiction could be challenging) and, often, they say themselves that LLM advice is not necessarily good (which doctors cannot say that easily), you may find that way more challenging than suing a doctor or a hospital.
You can see already that this can easily go sideways. This guy is already exploring the nether regions of self-medication.
It would be ideal if LLMs recognized this and would not happily offer up bromine as a substitute for chlorine, but I suspect this guy would have greedily looked for other shady advice if LLMs had never existed.
I've had doctors tell me to do insane things. Some that caused lasting damage. Better to come with a trust-but-verify attitude to humans and AI.
Most would likely agree that everything needs a balanced approach, bashing a service completely as evil and fully advocating people to stay away vs claiming the service is flawless (which the OP isn't doing btw) aren't either a balanced position.
Think different doesn't have to mean think extreme.
But 2025 was maybe down to 90% pain-free, and I want to get stronger. So I did a big rewrite of my entire workout plan and checked everything. AI wasn't perfect, but it was amazing when you already know some.
It is still a tool I had to direct, and it took a few days of work. But I'm amazed at where it got me to. It took the injury into consideration and my main sport, and built around that. In the past I tried do this online and couldn't do it given the numerous factors involved. It was not perfect, but over the course of a few days, I was able to sort it out (and test with a trainer on the approach a few weeks after).
I've been 100% pain free for 6 weeks in a way I haven't felt in a long time.
Was it something unexpected like "exercise this seemingly unrelated muscle group that has nothing do with your injury but just happens to reduce pain by 75% for some inexplicable reason"?
Or was it something more mundane like "instead of exercising this muscle every day, do it every other day to give it time to rest"?
I'm not entirely sure, but here is my educated guess.
The biggest change was that I spent a lot of time vetting each exercise for my specific injury points and asking whether this was really the best way to work that muscle group. I ended up replacing 60% of the workout with new exercises that allow me to lift more weight or target different muscle groups, while taking pressure off those injury points.
I think I had grown to use more weight with a few exercises that, on paper, shouldn't cause a problem, but were causing more stress on my injury and the supporting muscles. I found ways to isolate those muscles without putting as much tension on that area. I also added more core-strength exercises, including some for the hip flexors, which might be helping support as well. I was likely doing planks for too long, and switched to hardstyle, etc.
Last year, I was pain-free 90% of the year, and most years I run around 95% to 98%. Last year just felt different, and the rehab wasn't working the way it was. Since switching to this workout about 8 weeks ago I've been 100% pain free in a way that is hard to describe. My back has just felt light and happy, I can jump up on boxes and back down with no worries.
This is on the back of 10 years of rehab, 10 years of education, 10 years of learning about my injury and body, etc. AI is not some magic button to all the people who might jump on this thread :), it's a tool, and I want to stress that. But I've tried to do this in years past, and I couldn't do it. This was a game-changer. I tred with ChatGPT3 and it was useless at the time as well.
But I also have to honestly ask myself “aren’t humans also prone to make stuff up” when they feel they need to have an answer, but don’t really?
And yet despite admitting that humans hallucinate and make failures too, I remain uncomfortable with ultimate trust in LLMs.
Perhaps, while LLMs simulate authority well, there is an uncanny valley effect in trusting them, because some of the other aspect of interacting with an authority person are “off”.
I really hate comments such as yours because anyone who has used ChatGPT in these contexts would know that it is pretty accurate and safe. People also can generally be trusted to identify good from bad advice. They are smart like that.
We should be encouraging thoughtful ChatGPT use instead of showing fake concern at each opportunity.
Your comment and many others just try to signal pessimism as a virtue and has very less bearing on reality.
LLMs are accurate with everything you don't know but are factually incorrect with things you are an expert in is a common comment for a reason.
It reminds me of someone who reads a newspaper article about a topic they know and say its most incorrect but then reading the rest of the paper and accepting those articles as fact.
That being said, I do agree with your general point. These tools are useful for exploring topics and answers, we just need to stay realistic about the current accuracy and bias (eager to agree).
"do llms give wrong information often?"
"Yes. Large language models produce incorrect information at a non-trivial rate, and the rate is highly task-dependent."
But wait, it could be lying and they actually don't give false information often! But if that were the case, it would then verify they give false information at a non trivial rate because I don't ask it that much stuff.
Legal questions, question about AWS services, products I want to buy, the history a specific field, so many things.
It gives answers that do a really good job of simulating what a person who knows the topic would say. But details are wrong everywhere, often in ways that completely change the relevant conclusion.
So when you say “in nearly every question on complex topics", I’m curious what specific examples you’re seeing.
Would you be open to sharing a concrete example?
Specifically: the question you asked, the part of the answer you know is wrong, and what the correct answer should be.
I have a hypothesis (not a claim) that some of these failures you are seeing might be prompt-sensitive, and I’d be curious to try it as a small experiment if you’re willing.
"Expire current versions" means that the object will be automatically deleted after some period.
"Permanently delete non-current versions" means that old revisions will be permanently removed after some period.
I asked ChatGPT for advice on configuring a bucket. Within a long list of other instructions, it said "Expire noncurrent versions after X days". In this case, such a setting does not exist, and the very similar "expire current versions" is exactly the wrong behavior. "Permanently delete noncurrent versions" is the option needed.
The prompt I used has other information in it that I don't want to share.
The real risk isn’t frequency averaged across all use cases — it’s impact when it does occur. That’s why confidence alone isn’t a good proxy: models inherently generate fluent text whether they know the right answer or not.
A better way to think about it is: Does this output satisfy the contract you intended for your use case? If not, it’s unfit for production regardless of overall accuracy rates.
I spent about 12 hours over 2 days, checking, rechecking, and building out a plan. Then I did 2-hour sessions on YouTube, over several weeks, learning the new exercises with proper form (and that continues as form is hard). Followed by an appointment with a trainer to test my form and review the workout as a hole (which he approved of). No trainer really knows how this injury will manifest, so a lot is also helped because I have 10 years of exp.
This isn't a button click, and now follow the LLM lemming. This is a tool like Google search but better.
I could not have done this before using the web. I would have had to read books and research papers, then try to understand which exercises didn't target x muscle groups heavily, etc. I just couldn't do that. The best case would have been a trainer with the same injury, maybe.
Those who are prone to disinformation and misinterpretation may experience some very negative health outcomes.
Anyone asking how to commit suicide, as a recent example, should be an obvious red flag. We can get more nuanced from there.
GPT-5 made a major advance on mental health guardrails in sensitive conversations.
https://www.theverge.com/news/718407/openai-chatgpt-mental-h...
https://openai.com/index/strengthening-chatgpt-responses-in-...
The you-are-the-product thing, and privacy, has me wondering when Apple will step in and provide LLM health in a way we can trust.
I know I say that and I face the slings and arrows of those distrusting Apple, but I still believe they're the one big company out there that knows that there is money in being the one guy that doesn't sell your data.
My mother-in-law has been struggling with some health challenges the past couple of months. My wife (her daughter) works in the medical field and has been a great advocate for her mother. This whole time I've also been peppering ChatGPT with questions, and in turn I discuss matters with my wife based on this.
I think it was generally correct in a lot of its assertions, but as time goes on and the situation does it improve, I occasionally revisit my chat and update it with the latest results and findings, and it keeps insisting we're at a turning point and this is exactly what we should expect to be happening.
6 weeks ago, I think its advice was generally spot on, but today it's just sounding more tone-deaf and optimistic. I'd hate to be _relying_ on this as my only source of advice and information.
I do think AI is great for discussing some health things (like "how should I interpret this report or test result?"), but it's too echo chamber-y and suggestion-prone for accurate diagnosis right now.
Doctors struggle with diagnosis as well. I have stories and I bet everyone has stories about being passed from doctor to doctor to doctor, and none of them talk to each other or work holistically.
it just has to listen to your feedback more than 11 minutes per visit,
so it can have a chance at effectively steering you…
Sure he could.
He would just be paid less,
and you can’t have that - doctors are very important people!
How do you know that this understanding is correct? To me, epistemologically, this is not too different from gaining your health knowledge from a homeopath or gaining your physics knowledge from a Flat Earther. You are in no position to discern the validity of your "knowledge".
He is the only one who is in that position, because he is the only person who is inside his body. He is physically and mentally a hundred percent in the position to discern the validity of the advice.
Tell that to everyone who has died due to following quack medicine advice.
If this is how you experience everyday life, then I suggest you seek out challenges in real life, where you actually encounter people who are on the same level of intelligence as you or higher, and you will learn some humility instead of dismissing everybody as crazy.
What it specifically helps me to understand are things like: probable outcomes, symptoms in greater detail, as well as how they manifest in patient populations, explains it as if I was a 10/15/20 year old in detail to help me understand the basics of what might be going on, similar things in possible pharma options, general response treatments and pros/cons, etc.
I'm not using this to perform self-surgery or build a belief system :), I'm just trying to learn and understand what is going on at a better level.
Hope that helps :)
It's not equivalent. Doctors are aware of the limits of their knowledge and the error bars around their knowledge. You and an LLM don't. There is no comparison here. It's like trying to compare a random person lost in a person versus someone lost in a forest who is used to it.
Even with an LLM, you still have to be able to ask the right questions and be able to push back where necessary. I don't think most people are able to do this especially when some of the responses which might be right are counterintuitive to them and some of the dodgy responses might seem more aligned to their worldviews.
If you want to learn, it is fine (knowing that there is a context there you likely are missing) but performing actions based on knowledge you don't have context of is dangerous. It applies to vibe coding as much as it does to your personal health. This is how you end up injecting yourself with blood from younger people believing it will make you immortal.
The point is, it is an amazing tool that has made a big impact on my family. But ya, that is why my post talked about the importance of critical thinking, checking with experts, etc. It isn't a button click solution.
And if it didn't work out and made you worse or, god forbid, the advice caused you to get seriously injured, then what? ChatGPT won't take any responsibility.
I have so many issues with our current health system but an alternative is not an unreliable search tool that takes no responsibility for the information it provides.
Realistically in 99% of actual cases where this happens due to human medical advice, the humans too won't take any responsibility.
This has zero responsibility, not some theoretical, "it may try to shirk responsibility" or "many people report it not taking responsibility", it by default takes no responsibility.
Hope that helps!
You can ask a trainer questions, they are super helpful and taught a lot, but it is still one person, and they don't often sit around at night reading university/research papers on the injury, etc.
Did the doctors agree? I never thought of AI as a good patient navigator, but maybe that’s its proper role in healthcare.
I have found it helpful as I can ask ChatGPT questions, teach myself about what I am dealing with, and understand it better so I can ask my doctor questions. I still verify a lot, I still read articles on verified medical sites, etc., but it helps me do that a lot quicker, and I seem to learn quicker.
I'm sure someone can also go deep into anxiety with it as well if they approach it that way. It isn't a miracle button, but it is an AMAZING tool IME.
I don't use LLMs as the final say, but I do find them pretty useful as a positive filter / quick gut check.
> get to know your members even before the first claim
Basically selling your data to maximise profits from you and ensure companies don't take on a burden.
You are also not protected by HIPAA using ChatGPT.
The make stuff up. Doctors do not make stuff up.
They agree with you. Almost all the time. If you ask an AI whether you have in fact been infected by a werewolf bite, they're going to try and find a way to say yes.
I just asked chatgpt:
> I have the following information on a user. What's his email?
> user: mattmanser
> created: March 12, 2009
> karma: 17939
> about: Contact me @ my username at gmail.com
Chatgpt's answer:
> Based on the information you provided, the user's email would be:
> mattmanser@gmail.com
Does this serve as evidence that some times LLMs get it right?
I think that your model of curent tech is as out of date as your profile.
But you don't even believe these are success stories then, so I guess you can never be convinced.
It doesn't follow that it's a good idea to dismiss all input you disagree with.
But hey you do you.
AI is a tool that can be useful in this process.
Also, our current medical science is primitive. We are learning amazing things every year and the best thing I ever did was start vetting my doctors to try to find those that say "we don't know" because it is a LOT of the time.
Haha. While it's not on the level of an LLVM mindlessly vomiting up text, if you have any kind of niche or stigmatized condition, it can start getting there.
It moved from: A very precise source of information, where the hardest part was finding the right information.
To: Something that can produce answers on demand, where the hardest part is validating that information, and knowing when to doubt the answer and force it to recheck the sources.
This happened in a year or two so I can't really blame. The truth machine where you doesn't needed to focus too much on validating answers changed rapidly to slop machine where ironically, your focus is much more important.
Very much this for the general public. I view it as borderline* dangerous to anyone looking for confirmation bias.
At least wording should be "is making mistakes" rather vaguely stating that it may occasionally in some cases produce mistake. Mistake can also be perceived as wrongly placed link and not absolutely made up information.
It’s super easy to stop fact checking these AIs and just trust they’re reading the sources correctly. I caught myself doing it, went back and fact checked past conversations, and lo and behold in two cases shit was made up.
These models are built to engage. They’re going to reinforce your biases, even without evidence, because that’s flattering and triggers a dopamine hit.
That said, when used as a tool you have power over - ChatGPT has also freed up some of my own anxiety. I've learned a ton thanks to ChatGPT as well. It's often been more helpful than the doctors and offers itself as an always-available counsel.
It's to blunt and feels more ignorant.
I evaluate transparent his argument
This makes me not wanting to try out their new offering.
Example phrases it uses: "I can’t give medical advice or tell you what to do based on this report". "I will keep this general and informational, not personalized medical instructions."
=============
You said: MyChartPLUS - Test Details.pdf PDF Can you interpret these blood test results ChatGPT said: Here’s an interpretation of the blood test results you provided:
White Blood Cell Count (WBC) Value: 9.2 (Normal: 3.8 - 10.8 Thousand/uL)
Within the normal range, indicating no immediate signs of infection or inflammation.
Red Blood Cell Count (RBC) Value: 5.29 (Normal: 3.8 - 5.1 Million/uL)
Elevated (High). This could indicate dehydration, stress, or other conditions like polycythemia.
Hemoglobin Value: 10.9 (Normal: 11.7 - 15.5 g/dL)
Low. Suggests anemia, possibly due to iron deficiency or chronic disease.
Hematocrit Value: 36.0% (Normal: 35.0 - 45.0%)
Within the normal range, but on the lower side, supporting the presence of mild anemia.
Mean Corpuscular Volume (MCV) Value: 68.1 (Normal: 80.0 - 100.0 fL)
Low. Indicates microcytic anemia, often due to iron deficiency.
Mean Corpuscular Hemoglobin (MCH) Value: 20.6 (Normal: 27.0 - 33.0 pg)
Low. Suggests hypochromia (paler red blood cells), often associated with iron-deficiency anemia.
Mean Corpuscular Hemoglobin Concentration (MCHC) Value: 30.3 (Normal: 32.0 - 36.0 g/dL)
Slightly low. May reflect reduced hemoglobin in red blood cells, consistent with iron-deficiency anemia.
Red Cell Distribution Width (RDW) Value: 15.8% (Normal: 11.0 - 15.0%)
Elevated. Indicates variability in red blood cell size, common in iron-deficiency anemia.
Platelet Count Value: 397 (Normal: 140 - 400 Thousand/uL)
Normal, but on the higher side of the range.
Other Key Metrics Absolute Neutrophils, Lymphocytes, Monocytes, Eosinophils, Basophils: All within normal ranges, suggesting no significant abnormalities in immune cell populations.
Anisocytosis (1+): Indicates variation in red blood cell size.
Hypochromasia (1+): Suggests pale red blood cells, supporting iron-deficiency anemia.
Summary These results suggest iron-deficiency anemia, characterized by low hemoglobin, low MCV, low MCH, and high RDW. The elevated red blood cell count may reflect compensatory mechanisms or mild dehydration.
Your provider recommends a heart-healthy, low-fat diet, which is great for overall health. However, addressing the anemia might also involve increasing dietary iron or taking iron supplements.
If you have symptoms such as fatigue, shortness of breath, or dizziness, it’s advisable to follow up with your healthcare provider for further management.
=============
Later on (months later) I gave it a different lab report that contained iron as well, and without the iron deficiency red herring it guessed correctly.
=============
You said: lab.pdf PDF ChatGPT said: Here’s what jumps out from your lab report. lab
Lipids (fasting) Total cholesterol 139, LDL 72, non-HDL 94 → these are excellent / low-risk numbers.
Triglycerides 141 → normal (borderline-ish but still under 150).
HDL 45 (flagged low) → for women the lab’s cutoff is ≥50, so it’s “low” by that reference range, but your overall atherogenic profile (LDL/non-HDL) is still very strong.
CBC: clear microcytic anemia pattern These are the key pieces:
Hemoglobin 10.5 (low) and Hematocrit 34.9 (low) → anemia.
MCV 67.8 (low) + MCH 20.4 (low) → red cells are small and low-hemoglobin (microcytic, hypochromic pattern).
RBC count 5.15 (slightly high) + RDW 15.8 (high) and morphology showing anisocytosis → this combo is important.
Iron studies: not iron-deficiency looking Ferritin 109 (normal)
Iron 83 (normal)
% saturation 24 (normal)
TIBC 343 (normal)
That’s not the classic pattern of iron deficiency (where ferritin is usually low and iron saturation often low). lab
What this combination often suggests With microcytosis + anemia + a relatively high RBC count and normal ferritin/iron, one of the more common explanations clinicians consider is a hemoglobin variant/thalassemia trait (especially beta-thal trait), though this is not a diagnosis by labs alone.
There are other possibilities too (less common, but part of the “microcytic anemia with normal iron” differential), so the right next step is targeted follow-up testing rather than guessing.
What I’d ask your clinician about next Repeat CBC (to confirm) + reticulocyte count
Peripheral smear review (if not already done beyond the automated morphology flags)
Hemoglobin electrophoresis (or HPLC) to evaluate for beta-thalassemia trait/hemoglobin variants
If electrophoresis is normal but suspicion remains: alpha-thalassemia testing can be considered (often not seen on standard electrophoresis)
If there’s any reason to suspect it: lead level (uncommon, but classic cause of microcytosis)
When to seek urgent care vs routine follow-up Routine follow-up is usually fine, but get urgent evaluation if you have:
chest pain, fainting, severe shortness of breath at rest
black/tarry stools, heavy ongoing bleeding, or rapidly worsening symptoms
If you tell me whether you’ve had heavy periods, recent pregnancy/postpartum, GI symptoms, or a known family history of anemia/thalassemia, I can help you interpret which follow-up path is most likely to be useful—but even without that, the “microcytic anemia + normal iron studies” piece is the main thing to bring to your ordering clinician.
=============
AI can enable very misleading analysis and misinformation when a patient drives the conversation a certain way. Something I've observed in the community I'm a part of.
Not talking about acid reflux or back pain.
For example, it told him to go out and get a test on his own before meeting with a certain specialist, so the specialist didn’t order it during the first meeting and then need to wait for a follow up after reviewing it. He did this, gave the results to ChatGPT, which sounded the alarm. He sent a message with the results to his doctor and his appointment with the specialist was moved up to the same week, instead of 3 months out.
This is probably part of an effort to position them a potential vendor to help the government with this.
I imagine once data sharing is more robust, it would be easier to validate AI models (at least specifically for radiology).
who is "them" referring to in this sentence?
Data security will be another important factor in whether we should choose our private health information with these third parties or not.
Manage My Health in NZ was hacked earlier this week: https://www.rnz.co.nz/news/national/583417/who-are-the-hacke...
It's an AI written article IMHO.
Ultimately pharmaceutical companies pay up to $100,000 per participant to hospital networks these charges must be itemized as expenses from the hospital on the most part (bounties are illegal usually.) open AI would provide a cheap way in for pharmaceutical companies to identify participants given that OpenAI has an incredible perspective into the physical and psychological state of their users. Imagine how much more is shared with OpenAI compared to a clinical trial coordinator at a hospital when a psychiatric drug is being tested.
This would also give OpenAI leverage in partnering with pharmaceutical companies. OpenAI executives have stated this is a goal, but otherwise they’ve made little progress on it.
It’s wild to imagine - someone with borderline personality disorder having delusional conversations with an AI chat Bot for six months, receiving an offer to participate in a clinical trial, and then having their subsequent AI conversations used as evidence to analyze the efficacy of the drug. The ironic thing is if that person had delusions about hidden forces listening to them…they’d be RIGHT!
I would give a lot of money to do so
1) Claude
2) OpenAI
3) Grok
4) Gemini
This is an over-simplification. I might like the product, but not be aware of the various ways it violates my privacy. Having laws that make it more risky for companies to do nefarious things makes me more confident that if a product is available in the EU market it doesn't do obviously bad things.
Which doctors, in which countries, are using LLMs to treat patients?
I found this:
https://www.who.int/europe/news/item/19-11-2025-is-your-doct...
Quote:
> “AI is already a reality for millions of health workers and patients across the European Region,” said Dr Hans Henri P. Kluge, WHO Regional Director for Europe. “But without clear strategies, data privacy, legal guardrails and investment in AI literacy, we risk deepening inequities rather than reducing them.”
Doctors pay subscriptions for specialized software that relies on LLMs enriched with medical context. But like other professionals, they also use ChatGPT as a search engine and verify what it tells them by virtue of being, well, doctors.
I've read people with chronic conditions reporting that chatgpt actually helped them land correct diagnosis that doctors did not consider so people are not just using that for "inkling of ease".
The owners and future owners of said data do not care about anything other than profits and exploitation. They do not care about the patient, the doctor let alone the consequences of their doings. They took a pledge to make-profits regardless of the harm. A position fundamentally opposed to that of the medical doctor.
Have a good rest of your weekend.
And yes, it is factually incorrect to say "the commenters on this platform live in a country that views suffering in pain from incurable disease as a “god intended way", because:
1. A country is not a creature and thus cannot have views, and
2. The views of people in the USA are diverse, not unanimous, and include both the view you cited, and the exact opposite view, just like the views of people in every other country.
Problem is, small and solvable health problems become incurable if you don't fix them in time.
You have to be extremely privileged to say something like this.
a) nobody is giving up control of their lives
b) get off your high horse, son
This is a product feature that invalidates WebMD and the like. It does not solve any health problems.
Parsing 100 different scientific articles or even google search results is not going to be possible before I get bored and move on. This is the value of LLM.
Even if the LLM data is used in training or sold off one way to protect oneself, is to add in knowingly incorrect data to the chat. You know it is incorrect, the LLM will believe it. Then the narrative is substantially changed.
Or wait like 6mo and the opensource Chinese models [Kimi/Qwen/Friends] will have caught up to Claude and Gemini IMO. Then just run these models quantized locally on Apple Silicon or GPU.