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.
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.
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.
>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.
Sure, they were evolved using criteria based on next token prediction. But you were also evolved, only using critera for higher reproduction.
So are you really thinking, or just trying to reproduce?
You are implicitly assuming that no statistical model acting on next-token prediction can, conditional on context, replicate all of the outputs that a human would give. This is a provably false claim, mathematically speaking, as human output under these assumptions would satisfy the conditions of Kolmogorov existence.
However, the status quo is that "AI" doesn't exist, computers only ever do exactly what they are programmed to do, and "thinking/reasoning" wasn't on the table.
I am not the one that needs to disprove the status quo.
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.
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.
That's kind of like taking driving lessons in order to fix an engine. 'Early childhood development' is an emergent property of what could be cumulatively called a data set (everything the child has been exposed to).
ECD includes the mechanisms by which children naturally explore the world and grow.
I’m going to give you a spoiler and tell you that children are wired to explore and attempt to reason from birth.
So to fix your analogy, you reading about ECD is like you learning what an engine is before you tell a room full of people about what it does.
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.
> 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.
still am, unless i re-interpret "do" as "due".
“Area man who had poor judgement ten years ago now has both poor judgement and access to chatbots”
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...
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.
Your comment gave me a nightmare of that returning, but in AI form somehow.
Sure the machine very confidently lies about dangerous things, but you shouldn't trust it. But you should employ it to replace humans.
> 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?
some_random•1h ago
>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•1h ago
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.
zahlman•1h ago
Does ChatGPT ever ask the user, like, anything?
fl7305•52m ago
Yes. At least when I just tried ChatGPT-5:
Can I replace sodium chloride with sodium bromide?
ChatGPT said: Yes, in some cases — but it depends on the application.
Chemistry/lab use: Both are salts of sodium and dissolve similarly, but bromide is more reactive in some contexts and heavier. It can change reaction outcomes, especially in halide-sensitive reactions (e.g., silver halide precipitation).
Food use: No — sodium bromide is toxic and not approved as a food additive.
Industrial processes: Sometimes interchangeable (e.g., certain brines, drilling fluids) if bromide’s different density, solubility, and cost are acceptable.
What’s your intended use?
OJFord•1h ago
kragen•1h ago
> 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.
OJFord•25m ago
Would be interesting if he started to become symptomatic and so asked ChatGPT and that's where he got the idea that it needed to be replaced with something though. (But I suspect it was more along the lines of salty taste without the NaCl intake.)
topaz0•1h ago
zahlman•1h ago
> 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.