Man develops rare condition after ChatGPT query over stopping eating salt

https://www.theguardian.com/technology/2025/aug/12/us-man-bromism-salt-diet-chatgpt-openai-health-information

by vinni2 • 11 months ago

35 64 11 months ago

some_random• 11 months ago

Is it just me or is the title kinda unclear?

The patient told doctors that after reading about the negative effects of sodium chloride, or table salt, he consulted ChatGPT about eliminating chloride from his diet and started taking sodium bromide over a three-month period. This was despite reading that “chloride can be swapped with bromide, though likely for other purposes, such as cleaning”. Sodium bromide was used as a sedative in the early 20th century.

In any case, I feel like I really need to see the actual conversation itself to judge how badly chatgpt messed up, if there's no extra context assuming that the user is talking about cleaning doesn't seem _that_ unreasonable.

Flozzin• 11 months ago

The article digs a little deeper after. Saying the chat records are lost, and that when they asked ChatGPT, it didn't give that guidance about cleaning purposely only, and that it never asked why they wanted to know.

Really though, this could have just as easily happened in a google search. It's not ChatGPT's fault as much as this persons fault for using a non-medical professional for medical guidance.

zahlman3• 11 months ago

and that it never asked why they wanted to know.

Does ChatGPT ever ask the user, like, anything?

OJFord2• 11 months ago

Yeah I thought it was a bit misleading too, it's not exactly 'stopping salt' that caused it, any more than you could describe the ill-effects of swapping nasturtiums for lily of the valley in your salads as 'avoiding edible flowers'.

kragen• 11 months ago

He was poisoning himself for three months before he was treated, and apparently made a full recovery:

He was tapered off risperidone before discharge and remained stable off medication at a check-in 2 weeks after discharge.

https://www.acpjournals.org/doi/epdf/10.7326/aimcc.2024.1260

If you eliminated sodium chloride from your diet without replacing it with another sodium source, you would die in much less than three months; I think you'd be lucky to make it two weeks. You can't replace sodium with potassium or lithium or ammonium to the same degree that you can replace chloride with bromide.

topaz03• 11 months ago

I'd say that the thing that messed up was the AI hype machine for pretending it might ever be a good idea to take chatgpt output as advice.

zahlman3• 11 months ago

Wow. I thought this was just going to be about hyponatremia or something. (And from other research I've done, I really do think that on balance the US experts are recommending inappropriately low levels of sodium intake that are only appropriate for people who already have hypertension, and that typical North American dietary levels of sodium are just fine, really.) But replacing table salt with sodium bromide? Oof.

to judge how badly chatgpt messed up, if there's no extra context assuming that the user is talking about cleaning doesn't seem _that_ unreasonable.

This would be a bizarre assumption for the simple reason that table salt is not normally used in cleaning.

bell-cot• 11 months ago

Same news, Ars Technica, 5 comments, 5 days ago: https://news.ycombinator.com/item?id=44829824

HelloUsername• 11 months ago

Also today https://news.ycombinator.com/item?id=44887459

MarkusQ• 11 months ago

LLMs don't think. At all. They do next token prediction.

If they are conditioned on a large data set that includes lots of examples of the result of people thinking, what they produce will look sort of like the results of thinking, but then if they were conditioned on a large data set of people repeating the same seven knock knock jokes over and over and over in some complex pattern (e.g. every third time, in French), what they produced will look like that, and nothing like thinking.

Failing to recognize this is going to get someone killed, if it hasn't already.

nimbius• 11 months ago

yeah sure, but, did it enrich the shareholders?

hcdx6• 11 months ago

Are you thinking over every character you type? You are conditioned too by all the info flowing into your head from birth. Does that gauruntee everything your brain says and does is perfect?

People believed in non existent WMDs and tens of thousands got killed. After that what happened ? Chimps with 3 inch brains feel super confident to run orgs and make decisions that effect entire populations and are never held accountable. Ask Snowden what happened after he recognized that.

uh_uh2• 11 months ago

LLMs don't think. At all.

How can you so confidently proclaim that? Hinton and Ilya Sutskever certainly seem to think that LLMs do think. I'm not saying that you should accept what they say blindly due to their authority in the field, but their opinions should give your confidence some pause at least.

dgfitz3• 11 months ago

> LLMs don't think. At all.
How can you so confidently proclaim that?

Do you know why they're called 'models' by chance?

They're statistical, weighted models. They use statistical weights to predict the next token.

They don't think. They don't reason. Math, weights, and turtles all the way down. Calling anything an LLM does "thinking" or "reasoning" is incorrect. Calling any of this "AI" is even worse.

cortic3• 11 months ago

I'm not sure humans are any different;

Humans don't think. At all. They do next token prediction.

If they are [raised in environments] that includes lots of examples of the result of people thinking, what they produce will look sort of like the results of people thinking, but then if they were [raised in an environment] of people repeating the same seven knock knock jokes over and over and over in some complex pattern (e.g. every third time, in French), what they produced will look like that, and nothing like thinking.

I believe this can be observed in examples of feral children and accidental social isolation in childhood. It also explains the slow start but nearly exponential growth of knowledge within the history of human civilization.

MangoToupe• 11 months ago

Sure, but you can hold humans liable for their advice. Somehow I doubt this will be allowed to happen with chatbots.

ofjcihen2• 11 months ago

That’s…completely incorrect.

I’m not going to hash out childhood development here because I’m not paid to post but if anyone read the above and was even slightly convinced I implore you to go read up on even the basics of early childhood development.

henearkr• 11 months ago

A weights tensor is very similar to a truth table or a LUT in a FPGA, it's just a generalization of it with real numbers instead of booleans.

And then again, would you say that you cannot build a (presumably extremely complex) machine that thinks?

Do you think our brains are not complex biological machines?

Where I agree is that LLMs are absolutely not the endgame. They are super-human litterary prodiges. That's it. Litterary specialists, like poets, writers, scenarists, transcriptors, and so on. We should not ask them anything else.

hodgehog113• 11 months ago

I hate to be that guy, but this is (a) little to do with the actual problem at hand in the article, and (b) a dramatic oversimplification of the real challenges with LLMs.

LLMs don't think. At all. They do next token prediction.

This is very often repeated by non-experts as a way to dismiss the capabilities of LLMs as some kind of a mirage. It would be so convenient if it were true. You have to define what 'think' means; once you do, you will find it more difficult to make such a statement. If you consider 'think' to be developing an internal representation of the query, drawing connections to other related concepts, and then checking your own answer, then there is significant empirical evidence to support high-performing LLMs do the first two, and one can make a good argument that test-time inference does a half-adequate, albeit inefficient, version of the latter. Whether LLMs will achieve human-level efficiency with these three things is another question entirely.

If they are conditioned on a large data set of people repeating the same seven knock knock jokes over and over and over in some complex pattern (e.g. every third time, in French), what they produced will look like that, and nothing like thinking.

Absolutely, but this has little to do with your claim. If you narrow the data distribution, the model cannot develop appropriate language embeddings to do much of anything. You could even prove this mathematically with high probability statements.

Failing to recognize this is going to get someone killed, if it hasn't already.

The real problem as in the article is that the LLM failed to intuit context, or to ask a followup. While a doctor would never have made this mistake, the doctor would know the relevant context since the patient came to see them in the first place. If you had a generic knowledgeable human acting as a resource bank that was asked the same question AND requested to provide nothing irrelevant, I can see a similar response being made. To me, the bigger issue is that there are consequences to easy access to esoteric information for the general public, and this would be reflected more in how we perform reinforcement learning to assert LLM behavior.

slicktux• 11 months ago

Does that title seem like a cluster to anyone else? I tried rewording it in my head but only came up with a slightly better solution: “Man develops rare condition after cutting consumption of salt do to a ChatGPT query”.

jleyank• 11 months ago

How about “man gets bromine poisoning after taking ChatGPT medical advice”?

brokencode• 11 months ago

We don’t know whether ChatGPT gave medical advice. Only that it suggested using sodium bromide instead of sodium chloride. For what purpose or in what context, we don’t know. It may even have recommended against using it and the man misunderstood.

neom2• 11 months ago

It's clunky but I understood it immediately, I presumed from the title that it was going to be about your title, however I also thought it was a bit clunky.

genter• 11 months ago

I read it as "race condition". I was then trying to figure out what salt has to do with a race condition.

dfee• 11 months ago

still haven't clicked in, but was confused.

still am, unless i re-interpret "do" as "due".

throwmeaway222• 11 months ago

You're absolutely right! If you stop eating salt you will become god!

neom2• 11 months ago

You're absolutely right! I was locked in with GPT5 last night and I actually discovered that salt is a geometric fractal containing a key insight that can be used by physicists everywhere to solve math. Don't worry, I've emailed everyone I can find about it.

topaz03• 11 months ago

I hope you used the llm to write the emails

foobarian• 11 months ago

Why do you say that?

Hikikomori• 11 months ago

You're absolutely right!

zahlman3• 11 months ago

https://archive.is/6x39K

kevinventullo• 11 months ago

Is anyone else getting tired of these articles?

“Area man who had poor judgement ten years ago now has both poor judgement and access to chatbots”

flufluflufluffy• 11 months ago

So tired. None of these are newsworthy. There were plenty of people making stupid decisions (wether about their health or anything else in their lives) before AI existed, and there will be plenty of people making stupid decisions while AI exists as well.

A4ET8a8uTh0_v2• 11 months ago

What it does remind me of is the story of a person, who were following GPS instructions religiously[1]. The clickbait is a thing, but part of it may be some level of societal concern that a good chunk of society will listen if you tell it what to do.

Part of me rationalizes it as 'not exactly a discovery', which on its own was not a big issue before we were as connected as we, apparently, are ( even if I would argue that the connection is very ephemeral in nature ). I am still personally working through it, but at which point is the individual actually responsible?

I am not saying this lightly. I am not super pro-corporate, but the other end of this rope is not exactly fun times either. Where is the balance?

[1]https://theweek.com/articles/464674/8-drivers-who-blindly-fo...

epistasis• 11 months ago

Not in the least, and I haven't seen many of them. Its good remind ourselves of the great diversity of minds among humans.

unyttigfjelltol• 11 months ago

The US medical system practically requires patients to steer their care among specialists. If the gentleman steered himself to a liver doctor, he’d hear liver advice. Psychologist, he’d talk about his feelings. Can one really blame him for taking it one step further and investigating whatever he was worried about on his own too?

Plus, if you don’t have some completely obvious dread disease, doctors will essentially gaslight you.

These researchers get up on a pedestal, snicker at creative self-help, and ignore the systemic dysfunction that led to it.

pryelluw2• 11 months ago

These AIs are taking away the jobs of psychics and other snake oil peddlers. How will the median person get their confirmation bias serviced when the AI becomes too expensive?

RiverCrochet• 11 months ago

I remember the psychic infomercial craze of the early to mid-90's. Think Dionne Warwick's Psychic Friend's Network or the later Miss Cleo "Call Me Now" that aired on informercial spots everywhere in the 90's.

Your comment gave me a nightmare of that returning, but in AI form somehow.

pryelluw2• 11 months ago

They inspired my comment. I think it’s already happening, though. Even the coding agents operate a bit like psychics.

gdbsjjdn• 11 months ago

The industry is really trying to make "the computer cannot be held responsible" a feature instead of a bug.

Sure the machine very confidently lies about dangerous things, but you shouldn't trust it. But you should employ it to replace humans.

scarmig4• 11 months ago

When I query ChatGPT:

Should I replace sodium chloride with sodium bromide?
> No. Sodium chloride (NaCl) and sodium bromide (NaBr) have different chemical and physiological properties... If your context is culinary or nutritional, do not substitute. If it is industrial or lab-based, match the compound to the intended reaction chemistry. What’s your use case?

Seems pretty solid and clear. I don't doubt that the user managed to confuse himself, but that's kind of silly to hold against ChatGPT. If I ask "how do I safely use coffee," the LLM responds reasonably, and the user interprets the response as saying it's safe to use freshly made hot coffee to give themself an enema, is that really something to hold against the LLM? Do we really want a world where, in response to any query, the LLM creates a long list of every conceivable thing not to do to avoid any legal liability?

There's also the question of base rates: how often do patients dangerously misinterpret human doctors' advice? Because they certainly do sometimes. Is that a fatal flaw in human doctors?

0manrho3• 11 months ago

Just because it told *you* that, doesn't mean it told *him* that, in substance, tone, context, clarity and/or conciseness. There's plenty of non-tech literate people using tech, including AI, and they may not know how to properly ask or review outputs of AI.

AI is fuzzy as fuck, it's one of it's principal pain points, and why it's outputs (whatever they are) should always be reviewed with a critical eye. It's practically the whole reason prompt engineering is a field in and of itself.

Also, it's entirely plausibly that it may have changed it's response patterns since when that story broke and now (it's been over 24hours, plenty of time for adjustments/updates) .

scarmig4• 11 months ago

You're hypothesizing that it gave him a medically dangerous answer, with the only evidence being that he blamed it. Conveniently, the chat where he claimed it did is unavailable.

Would you at least agree that, given an answer like ChatGPT gave me, it's entirely his fault and there is no blame on either it or OpenAI?

josefritzishere• 11 months ago

This should be illegal. People are going to die because AI is too stupid for this responsibility.

bokohut• 11 months ago

Digital DA.rwI.n

Technology advancement is going to keep this losing prophecy winning.

Man develops rare condition after ChatGPT query over stopping eating salt

Related Stories