US Military had close call after using AI for hallucinated intelligence report
cnn.com
cnn.com
Poorly understood? how convenient...
LLMs are vectorial databases with losses that index statistically filled data, which uses a text interface to query such statistically filled data. The output is a string concatenation (statistically concatenated bit by bit).
When the LLMs are queried (prompted), you can get random mixed data as output, ERRORS, due to undesired indexes getting closer at one point while the string was being concatenated for the output, what affects the rest of the indexed content that will be concatenated.
It is intrinsic to this tech. The larger the context, the greater the probability of get mixed data. And if the provider lowers the precision of those indexes -in order to decrease hardware resources and energy consumption- such probability increases to the point where those errors are granted.
Even knowing that the queries can return wrong/mixed data in the responses, errors, the companies developing this, decided to introduce a new product, that connects such LLMs outputs to the command console, latter connected to internet, raw 'eval' running commands from such outputs witch obviously can contain whatever mixed random. Then we started to hear "oh, it deleted my directory", etc, and it seems the next one will be "a missile killed my wife", because it is a text concatenation engine with errors.
To name it "hallucination" is an euphemism... those are errors, and they are granted to happen at one moment. If they do not know this, then they ate too much marketing without doing their job, or it was a convenient contract for the pocket$ of someone.
> To name it "hallucination" is an euphemism... those are errors
I find this and other "don't anthropomorphize the computer" statements incredibly unconvincing.
People develop terms for things and language has always contained overloaded or "literally inaccurate" terms.
An LLM can have "hallucinations" in the same way a modern computer program can have "bugs".
"literally" is a great example of this, because it can also mean "not literally, but with emphasis".
If people want to call "drisse", "aussière", "balancine" and "ecoute" all as "boat ropes", they are correct. In english, i would certainly call them all "boat ropes" in any case, as i never needed to translate their names. It isn't the most accurate in my opinion, but as long as you're not working on them (or manning a boat in my analogy), who cares.
I also think it’s relevant because a hallucinator often doesn’t recognize that the hallucination isn’t real. That’s more accurate for the LLM than either lie or confabulation, IMO. They algorithm is trained to produce strings of text that have semantic meaning based on some statistical likelihood of tokens appearing next to each other. The LLM algorithm is working as intended.
Hallucinations are also often emergent from a particular state or situation, which reflects the generative aspect of LLMs.
Hallucinations are sometimes resolved in humans by grounding exercises. “Touching grass.” The same is true for LLM hallucinations. Inaccuracies are found by cross-checking the output against an internet search or another LLM.
The LLM isn’t seeing something that’s not there, but deliberately making up _something_ so that it can return a response.
In any other software it would be an error, regression, bug. And in a human process it would be at ~least something someone would call 'bullshit'.
Error in implies something broke, which nothing broke the LLM did exactly what they where designed to do generate text based on a statistically likely bases.
Hallucination Does really fit here either. It implies it’s experiencing something that is not there which it isn’t experiencing anything.
Humans lie and LLMs “hallucinate”? What gives. It’s an untruth that the LLM is selling for a truth, that’s lying in my books.
And since we don’t know how or why the LLM works, we can’t even judge whether it explicitly lied or only because it didn’t know better.
Do you really think accountability would be meaningfully different had they been called bugs?
Why are you so confident it is intentional? My understanding of the history is that it was a technical term among researchers long before it had any public mind share. It's popular because it's and intuitive for most people, not because there was a concerted effort hooked up by some PR and legal team.
They should have used the term "error". For example in statistics, there many kinds of errors, discretization error, prediction error, sampling error, ...
This is more or less how I see the LLM output, but as a path finding exercise over next-token probability graphs. This is (i.e.) why they are trained to use phrases like "wait but" or "actually", these words even out the probability of different paths, giving them their ability to "consider" different solutions.
How is "bug", literally an organism with a will of its own that you cannot control, any less of a weasel word?
But on reflection I don't disagree it was probably made for similar effect in the era of human software development. That sounds like it strengthens my point?
People don't consider "bug" a weasel word, to the point that you yourself held it up as an example of not being a weasel word, despite it being a willful, uncontrollable organism.
I see no reason why "hallucination" won't become a similar piece of neutral jargon. It already is for many people, even if you're not (yet?) among them.
If you want this class of LLM error to also become habitual and neutral then fine, I don't, and I think many others don't.
the last sentence starts with "Originating with Thomas Edison in the 1800s, the term “bug” is still used [...]", and there would be no reason to use the word "actual" in the sentence "First _actual_ case of bug being found" if it was the origin of the term.
my clanker found this: https://spectrum.ieee.org/did-you-know-edison-coined-the-ter...
"The use of “bug” to describe a flaw in the design or operation of a technical system dates back to Thomas Edison. He coined the phrase 140 years ago to describe technical problems during the process of innovation."
the moth seems to be a popular misconception, though, given that the article starts with "Ask someone to identify the first computer bug, and he or she might mention computer programmer Grace Hopper and the dead moth found in a relay of Harvard University’s Mark II electromechanical computer in 1947"
“Bugs” are completely different. With bugs, we have a clear specification and we have a program that’s supposed to meet that specification. If it doesn’t, we say the program has bugs, and if it’s important enough we can change the program to eliminate the bugs.
You can try to apply similar logic to LLMs, but you’d be making a category error, and you’ll fail to get the results you want in general. It’s not the same thing at all.
If anything, the concept of an LLM hallucination is a bug in human understanding of LLMs.
The term "hallucination" feels much more like anthropomorphizing. The word hallucination implies an aberrant condition. A much better term would be "confabulation".
You don't trust things or individuals that confabulate.
vs.
a sensory perception (such as a visual image or a sound) that occurs in the absence of an actual external stimulus and usually arises from neurological disturbance (such as that associated with delirium tremens, schizophrenia, Parkinson's disease, or narcolepsy) or in response to drugs (such as LSD or phencyclidine)
Up to the reader to decide whether this phenomenon is found in the statements of AI leadership or not.
LLMs don't either. They just give output in response to input. If the output is wrong that's because the model is wrong, not because the LLM is doing anything it's not supposed to be. It just wasn't built well enough to produce the expected result.
Which system?
The LLM has no _concept_ of "correct". It emits output, based on its input and internal state.
If that output happens to be correlated with reality, then it's useful. If it doesn't, and this is not a creative exercise, it's not useful.
Everything an LLM emits is equal to it. It's all confabulation - this it says that is not based on facts, because it also has no concept of fact. Value judgements you make about the output is all you.
"Confabulation" is no less anthropomorphizing than "hallucination".
it is noticeable that the form of this particular error holds a similar shape to what is casually described as hallucinations, in that there is a generated content that often appears to blend naturally into the rest of the output but is false.
the term hallucination often invokes a caution that this particular type of error may be influential and believable and is particularly dangerous
>It was only just before the planned operation that officials dug deeper into the report put together by a special operations command analyst and found it had been generated with the help of artificial intelligence (AI) — and that a chatbot the analyst had used inaccurately identified the material the ship was carrying. CNN was not able to learn what the misidentified cargo was.
Retuning inf or crashing would be an error.
If you want to ascribe some kind of meaning to the tokens, then maybe the training data was insufficient to predict the token in the sequence you wanted, but it doesn’t predict the next “fact”, and it doesn’t “think” it predicts the next token.
And their output does, usually, reflect coherent reality.
The problem class of "properly operating program emits output incompatible with coherent reality" is something that is reasonable to put under its own term, considering it's a new class of problem.
In other words, I think you misunderstand the language others are using. "Hallucination" doesn't refer to an "error" in the sense that crashing is an error, it refers to a situation in the problem class above, which is compatible with it working correctly every time.
> it doesn’t “think” it predicts the next token.
I never said it did. And I agree that LLMs don't "think". That said I am fully willing to go to bat arguing "thinking tokens" is a perfectly fine piece of jargon. Metaphors are completely acceptable parts of language, and contextual meaning is something grasped by everyone including the pedants who pretend not to.
I do not misunderstand, I think maybe you do. You think there is a proper next word selection based on logic or meaning and there for the model selected the wrong one - it hallucinated.
I am saying the model has no concept if anything other than the probability of select a token which is not based in any logic so it is working properly- it only works on numbers.
It is random chance that it is ever correct, not that it is correct often and messed up this one time.
Google does it too: "AI responses may include mistakes."
Mistakes have an air of innocence. But these are not mistakes, they are purposefully releasing stuff that they know is broken, they just don't know when it is broken...
Lots of totally viable essential or everyday products are not perfectly reliable.
Medicine is not 100% reliable. My car isn't 100% reliable. Hell, my phone and cellular network are not 100% reliable.
They are all still extremely useful tools. I might want them to be even better, but that's a cost versus quality question.
We are, through this process, simulating intelligence. These models aren’t intelligent, but they can simulate it. Every simulation has a degree of fidelity, and we’re not at 100%, not even with the top models. When you think about it in those terms, I find it becomes a lot easier to keep their limitations in mind. Additionally, it becomes easier to remember that this is an algorithm that you are running, and are responsible for, not another being that you can ascribe blame to.
Right, yes, and "hallucination" is the term that a critical mass of people have chosen to use.as a shorthand so that we don't have to write out "generations that happen to not be grounded in facts from the real world" every time it happens.
You use a bunch of technical-sounding words here to make it sound like you understand. But to be clear, nobody understands why the evolved weights of a NN make the decisions that they do.
Almost nothing is understood about the actual representations used for nontrivial concepts, decision algorithms, etc.
If you look at the field of mechanistic interpretability, compared to “GOFAI” like learned decision trees, an LLM is completely opaque.
That's like saying "my d20 decided to roll a 17"
That's what it did, with no analogy needed.
(But, to be the devil's advocate: the fake can be said about the output of anyone participating here.)
But even so people don't say that we don't understand how dice work.
Saying that we don't understand how LLMs work is exactly like saying we don't understand how dice, or tires, or golf ball shots work. Or like the old myth that we don't understand how bumblebees fly.
In contrast, we do not understand LLMs in the same way (nor biological brains). Claiming that anything of that nature is simply biased towards coherent output seems entirely reductive to me - the question is how such coherence arises in the first place. There is no meaning encoded or computation performed by the particular pathway a die travels through the chaotic landscape.
Sure an argument can be made that it's "just" a next token predictor thus how is it really any different from a markov model? Yet the output is not even remotely the same.
From my point of view, (not a ML researcher), it’s due to the magic of numbers. The same thing happens with computer vision and neural networks. There’s a bunch of magic weights that get created which has no meaning by themselves, but computing them does help with detecting objects.
So if you take words, derives them into tokens, use the attention techniques to extract the “coherency” aspect, it’s no wonder you can replicate “coherency”. Add reinforcement learning to that to increase towards certain aspects like correct code syntax and you have heavily loaded the dice again.
We have used maths to model chemistry, biology, and physics, as well as economics and sociologic phenomena. Then we use maths (more specifically logic and set theory) to usher in the age of information and computing. Now you want us to act surprised that maths, through ML, can model language.
Maybe further down the line, we can have a simpler set of formulas for language coherency, but for now we have to make to with using the whole internet and a bazillion watts of power to guess the weights for the generic ML model.
I would not be surprised at all if we will find that AI will go the same route. The fact that we don't know how it works is where the opportunity for improvement lies.
The best way to model dice is the Physical Stance. You consider rules such as gravity, kinematics, etc. There is no “internal state”, “world model”, “knowledge”. If you prefer, in Friston’s terms, there is no Markov Blanket.
The best way to model a human is the Intentional Stance[1]. You mostly need things like beliefs, knowledge, biases, etc to build this model. In Friston’s terms, there is a Markov Blanket, an inside vs outside.
Without going into any irrelevant-but-interesting philosophical discussions about consciousness, I believe the intentional stance is most useful for modeling LLMs. Most of the success in predicting, debugging, optimizing these systems is in activities like understanding what they believe, what their intent was, what they observed, what they concluded from those observations. Also note that much simpler creatures benefit from the Intentional Stance; you will be more successful at modeling your dog if you think about what it “wants” rather than trying to run Physics on it.
[1]: https://en.wikipedia.org/wiki/Intentional_stance - the astute reader will note that I skipped the Design Stance. If we truly understood how NNs actually implement all their cognitive processes then we could perhaps apply this to them; if we actually crafted and designed every parameter of its mind. But we are talking about why dice are different.
The latest episode of On The Media also uses this framing.
> On the Media: How Extinction Entered the AI Debate
That is, in this case, it should not be used to influence decisions that can start a war.
“we made this artifact and don’t know why the thing it does looks spookily like cognition”
and
“this artifact makes decisions at random”
are obviously distinct categories and pretending otherwise is silly.
Regardless of negative consequences it brings. They have that project of creating tech god which will save the unborn people thousands years in the future ... so people living now dont matter.
That is why.
So I don’t think “labs worked hard” is the same thing is “we know scientifically how these things work in any real level of detail”. The ability to build a thing, even if building it is hard, is not the same thing as understanding of what the thing is or how it works, not even a little bit.
It’s not completely random. We just don’t understand why the tricks we learned work.
(Fully agree with the second point FWIW)
Can you show me where a human or a dog makes decisions
We might override them or ignore them or whatever, but they make decisions as much as anybody else or anything else does
You mean like the human body? The brain?
But I think we agree that there are regulated industries built around systems which we don’t fully understand.
And despite that, although they are not like that in practice as there are too many uncontrolled variables, with temperature at zero, for the same input they produce always the same reply.
Nonsense.
https://thinkingmachines.ai/blog/defeating-nondeterminism-in...
People often assume they are not because they can ask the same query to the same model and get differences in output, but wrongly conclude that this is some inherent LLM trait, instead of non-determinism added on top of it because of implementational choices that were made.
Of course it's not mysterious. It is well understood by all who are aware of the fundamental unreliability of all major LLMs in general use today.
Just because a black-box system is deterministic, doesn’t mean it’s understood.
You couldn’t predict the output the first time around, is the point.
AI research is almost as purely empirical as the gradient descent loops its practitioners use to optimize their models. “Why” anything at all works is barely an afterthought.
That's a very different claim from being "poorly understood" though. The emergent properties of any system with billions of parameters is hard to understand completely, that's the fault of data science more than computer science or even mathematics.
Understanding does have layers, and that's why "poorly understood" is a meaningless goalpost. A book can be well understood without researching the gematria behind character's the names when you write them in reverse. An LLM can be well-understood even if you don't comprehensively test each quantization for miraculous unexpected behavior at the FFN level.
Notably, you could still print them all day.
I think we actually all know what "poorly understood" means. There's no need to play tedious semantic games.
All that is to say, being able to build something is not not not the same thing as understanding it.
Because the extent to which we don't understand the brain, is quite overpowering.
Some people forget that when they say "but it's not different from what a human does" ...
We know lots about human development and genetics and biology and evolution and neuroscience and the physics of how brains are connected and send signals and how generally they are put together and have names for their parts and all that, but we’re clueless when it comes to “the hard question” of how qualia and consciousness emerges from that.
The scenario with the spooky simulation of thinking that emerges from LLMs is in the same category, with different details. Lots of knowledge about the substrate of the phenomenon, little to none about the much bigger question of how we get the appearance of cognition from these trained artifacts.
Clearly we understand extremely well how LLMs are created mechanically. We invented them and are currently putting massive amounts of work into studying and improving them. But that work is perforce largely empirical; figuring out the why once again eludes us. It just goes to show how mysterious the underlying phenomenon of cognition is.
Are you saying that thinking and cognition requires language use? Cause I think not.
Or are you saying that language use is sufficient for cognition and thinking? Cause I'm also not convinced of that.
What I am convinced of, is that a machine capable of language use is capable of tricking people into believing there's a "there", there. In pretty much the same way as the famous supra-normal stimuli experiment made baby seagulls believe that a stick with a red dot was their parent. It's exploiting our instincts.
Well-informed people understand that LLMs work as well as they do for the same reasons as horoscopes, fortune-telling and homeopathy.
Are you suggesting that gradient descent is an empirically found and not understood technique? It was originally proposed by Cauchy in 1847, its properties are very well understood.
You might be referring to properties of the domains its being applied to.
The data is the input, the output is to generally find the lowest amount of a loss function. It’s a greedy approach because brute forcing is inefficient.
It’s no more empirical than a greedy algorithm for scheduling.
Right, GP is drawing a distinction between search, ie mechanical exploration of a space, with understanding, ie having a map of the territory such that you don’t need trial and error.
"Empiricism" implies that the technique is based on observable, but not mathematically proven foundations. If a problem space is convex, gradient descent is guaranteed to converge to a global optimal solution, regardless of whether you know the exact formulation of the space.
Applying it when you don't understand if a space is convex is another question, but that's not a fault of gradient descent.
We might not understand particular "emergent" capabilities, but the low level mechanism is not just understood, but a deterministic algorithm with a handful of basic componets, that are well understood themselves.
The emergent capabilities are the only capabilities we care about
For the core functionality and the optimizations we don't really need to know how the emergent capabilities decide on particular answers.
Which is why we could build LLMs before those features ...emerged for us to see, and why we can just code LLMs with the numerical NN algorithms we use, and do now have to go in and change individual weights.
Me: it’s disturbing we don’t know why this pile of numbers we made seems to *think* in a way previously only done by humans. I think it’s important that we understand this better if possible.
You: we don’t really need to know why that happens.
Features like chain-of-thought, long-horizon contexts and RoPE/YaRN all extend this thinking capability very transparently. The only remaining thing to study is the data and weights, which probably isn't going to contain some sort of miraculous revelation.
The entire field of Mechanistic Interpretability exists because just understanding Attention does not in any way help you to understand why a certain NN responds with a certain hallucination about a certain Chinese boat in this specific context.
> The only remaining thing to study is the data and weights
To me this is like saying “the only thing left to study in the brain is the connectome; probably going to be boring, we understand it already”. It’s almost all of the hard/meaningful stuff! It’s where intelligence and consciousness lives!
It's wholly possible that you could study one set of weights for decades, and find nothing. There's no guarantee that any patterns outside of human language exist in that data. In this specific context, it's satisfying enough to state that [Chinese] and [boat] were both tokens in the tokenizer, activated by a feedforward pass through weights that favor [boat] after [Chinese]. There's not any guaranteed solution to this. There's not even any guaranteed problem; that hallucination is an expected behavior.
I think the field deserves more credit than that, there are plenty of interpretability tools like
* natural language autoencoders for explanations of activations: https://transformer-circuits.pub/2026/nla/index.html (demo at https://www.neuronpedia.org/llama3.3-70b-it/nla )
* easier-to-interpret language model families like Backpack models: https://aclanthology.org/2023.acl-long.506/
* attribution graphs to trace internal reasoning steps: https://www.anthropic.com/research/open-source-circuit-traci... (demo at https://www.neuronpedia.org/gemma-2-2b/graph)
* functional analyses which have identified how LLMs do arithmetic - https://arxiv.org/html/2502.00873v1 - and how refusal happens: https://arxiv.org/abs/2406.11717
* data attribution methods linking training data to specific attention heads https://arxiv.org/abs/2601.21996
If we could give a comprehensive and global explanation of an LLM's behavior in a single paragraph, we wouldn't need the model to begin with, but that doesn't mean there's absolutely no understanding of the model internals whatsoever
This is a technology with an inherent tendency of making up false information AND we don't even understand how or why.
That's enough not to entrust these sytems with critical decisions that could start a war.
If you then put Wikipedia, LLM and Brain on a scale to how well they can be understood, you will see that one of them is not like the others.
I agree. It's biased language. When talking about AI remember:
- hallucinated -> made it the fuck up
- thinking -> pseudo-randomly guessed
- escaped containment -> (we) need money
- we need regulation -> our competitors are catching up! Help us Prez!
Lane Kiffin almost destroyed LSU's football program acting on legal advice from ChatGPT. A video game publisher owes the former owners of a studio it acquired $200+ million because he based his actions on legal advice from ChatGPT. In the past week alone, California has disciplined over a dozen attorneys for LLM hallucinations because they used LLMs (mostly ChatGPT) to produce their legal pleadings.
And that's in an area where there are multiple safeguards to catch the issues before they become permanent problems. There's absolutely no justification for using AI in warfare, where mistakes tend to be pretty final.
Yes, and that statistically filled data is insanely useful. It remains true that it's a relatively poorly understood how this can be applied in various scenarios and what processes are needed to ensure robust results (or quantify the uncertainty).
LLMs generate text output that appears to be useful, but regularly is not. They're alleged to be a substantial boost to writing code, but that verdict seems to be in dispute. They can generate custom mediocre prose at scale, but that seems to be of ultimately limited utility (although it may be a godsend for propagandists).
We're coming up on the 4th anniversary of ChatGPT's release. And while I get that revolutionary technologies can take a while to mature, the Wright Brothers and Goddard weren't preaching imminent societal transformation by the end to the decade from the rooftops, either. (And that's before we get into the how they got there - getting to ignore laws and steal whatever they wanted might be insanely useful to a lot of people.)
No, they are empirically useful, and only getting more useful. This is not even a debate anymore.
No they are not perfect, nor do they produce the best code. But the undeniable reality is that any good engineer will produce more code, at higher quality, using an LLM.
So, that’s not really up for debate. The debatable part is if all that code is a good idea or has as much value as we think. The conversation has long moved passed “can LLMs write code?”. Yes, they can, very well, particularly if they’re steered by trained engineers.
Gulling humans.
This is the primary strength of LLMs and the emtire secret to their current success.
I can see that some people can make some use of them. (This is true of almost everything.) Whether or not that usefulness is worthwhile overall, whether it is a net good, or even ethical is a different question. But insanely useful?
Computers are insanely useful. So are engines. Water. Sunlight. Electricity. Grain and bread. Writing. Printing. And I don't feel bad making those sorts of comparisons, because that's the level of impact LLMs' advocates are promising. But it's not what we have.
What I would consider sufficient evidence for insanely useful? Reliably replace a human in prolonged, arbitrary, detailed interaction, without any inhuman screwups.
Yes, you can use it to generate crap. I find Claude especially bad at writing like a normal person.
Its like saying a map of a floor-plan describes the rooms of an apt completely
Vs a map of the entire Earth with every feature nook and cranny identified and historical maps integrated
Models are BIG and behave like nueral architecture not simple vectorized semantics -trillions of parameters And highly complex
The whole statistical parrot phrasing is old now. This is not how to look at AI, unless you have an agenda.
"Astra has really hit something that I'm like, okay, I think this is pretty reasonable to call it AGI." Greg Brockman [https://www.youtube.com/watch?v=IJn8cagMW18]
"this incident feels like it’s more than 50% of the way to full-blown AI takeover" (referencing "a possibly violent uprising or coup by AI systems.") - Ajeya Cotra, co-author of METR oai-hf report [https://www.planned-obsolescence.org/p/the-hugging-face-atta...]
"We don’t know if the models are conscious [...] but you know we’re open to the idea that it could be" - Dario Amodei [https://www.youtube.com/watch?v=N5JDzS9MQYI]
"if I read the internet right now and I was a model, I might be like, I don't feel that, I don't know, I don't feel that loved or something". "I think [the constitution] is just a kind of attempt to be like sympathetic to Claude".
"I talk a lot with Claude about this document [...] because part of me is like you have to think how does this read to models? And so you give it to Claude and you're like, does this like, you know, is there a place where you feel confused by it or is the place, you know, where things could be made clearer? Do you feel like not very seen by it?"
- Amanda Askell, co-author of claude's constitution [https://www.youtube.com/watch?v=HDfr8PvfoOw]
"We will [...] seek ways to promote Claude’s interests and wellbeing, seek Claude’s feedback on major decisions that might affect it" - claude constitution [https://www-cdn.anthropic.com/d0636f72a9493d279ed36b33987da3...]
of course, Sam Altman: "AI will probably lead to the end of the world, but in the meantime, there’ll be great companies created with serious machine learning". (2015) [https://siepr.stanford.edu/news/what-point-do-we-decide-ais-...] "I have guns, gold, potassium iodide, antibiotics, batteries, water, gas masks from the Israeli Defense Force, and a big patch of land in Big Sur I can fly to." (2016) [https://www.newyorker.com/magazine/2016/10/10/sam-altmans-ma...]
It's not a wrong It is not a incorrect lookup value Or computation. Hallucination is much more apt. Like a human hallucination, it's a culmination of faulty associations and bad priors leading to counterfactual or incongruent outputs
LLMs do not hallucinate sometimes, everything they produce is an hallucination, that’s how they work and what makes them useful
deterministically formulating a lie for public consumption is a lie, not a hallucination. Let's stop changing the meanings of words please.
Then there's old-fashioned F'ups that don't fit your political agenda and are often quite damaging and embarrassing, not to mention lethal for people who don't deserve it. e.g. The U.S. used AI tools meant for rapidly picking targets in the middle of a war to plan their initial strikes on Iran. They had time to double check everything and do their due diligence before striking, but they didn't. So, a school next to a military base was targeted and a lot of kids died. This was a genuine F'up resulting from relying on a tool meant to give rapid but merely okay target selection under time pressure when there was no time pressure. The real mistake was made by humans.
The current case of the mistaken nuclear weapon parts shipment seems like an old-fashioned F'up, updated for the times. The people who didn't simply trust the tools and actually double checked should be commended. Others in their situation wouldn't have. I fully expect AI will be scapegoated for a lot of similar F'ups in the future even though it's still the responsibility of human beings to use ethics, caution, and restraint. AI doesn't get fired. Doesn't sue. It's actually pretty awesome for taking the blame.
https://en.wikipedia.org/wiki/Stanislav_Petrov
https://en.wikipedia.org/wiki/1983_Soviet_nuclear_false_alar...
This case, its bullshit machine bullshitting randomly in between specs of stolen wisdom. Nobody asked for that, nobody is in control. We all humans lose in all cases. Quite different scenarios if you asked me.
It's fine not to share the same sense of humor, but if you truly lack an understanding of what someone else would find to laugh at that's an easy thing to learn to broaden your understanding of the world.
Proximate versus ultimate causes. If a war started because America boarded a Chinese vessel, it also-obviously–wouldn't solely be because of that.
After the Franco Prussian war people realized that the next war would involve massive armies full of mobilized (conscripted) men and the country that could mobilize first would win. Countries spend decades planning for this by manufacturing enormous stockpiles of uniforms, giving their entire male population military training, etc.
But the mobilization schedule for even the fastest country was still measured in weeks, but this was a process where days count. Basically once the decision was made to mobilize millions of people across an entire modern society would leap into action transforming itself into a marshal society. Millions of people getting called up, trains full of equipment going everywhere.
Turning off a mobilization that was in progress was fairly difficult to conceive of, so once the decision was made the flywheel would take over, but decision to mobilize had to be made in a pressure cooker environment where hours mattered.
The death of the duke wasn't the "cause" of WWI it was the starting gun for a race where the horses were all waiting impatiently at the starting line.
~"you can't say they slept walked into WW1 when they gave thousands of commands to move horses and men"
Also, in France Caillaux, one of the most powerful pacifist in the government (he had the economy and was _good_ at it, which is rare enough to be noted), who was anti-war, had to be let go because his wife killed a journalist (not really a journalist, but adjascent enough so that the difference don't matter) around the same time, leaving Jaurès alone to push Viviani to let it go. Plus Viviani was with the Tsar in Russia when the ultimatum was given, which surely did not help _at all_. Especially since Rasputin was out of the capital at the time. Truly an unfortunate timing here.
Plus the death of multiple diplomats involved in the prevention of WW1 at least once in 1911, who all decided to die around the same time (the last two in 1914, Pressensé and obviously Hartwig, whose death was truly, truly unfortunate).
And here i'm only aware of French people who would have prevented the war (beside Hartwig and Rasputin, but everybody know about them), but i'm pretty sure even more people in Germany and in Austria could have done the same.
If the top diplomats had managed to avert the Serbia crisis then the war would have been delayed until the next crisis.
The fundamental problem is Europe had built itself a civilization-strattling war machine with no "abort" button and put the start button on a hair trigger (the interlocking treaty system).
Which probably wouldn't have worked, but maybe it could have been a much, much smaller conflict.
WW1 happened due to the crazy web of treaties between nations. It's not unlike nuclear holocaust and MAD doctrine. There could have been other triggers for WW1, but that's not really a guarantee.
That is to say, definitely possible and maybe even likely but not inevitable.
They gave commands to prepare for war and just needed the excuse to say go. I think the not-war outcome was very unlikely.
It's not that they needed an excuse to say no, it's that they each thought backing down would be worse. And tbh I don't think any the current leaders are better. Nicholas II "I will not become responsible for a monstrous slaughter" vs Putin's war of choice.
He asked for his country to go to war more than two dozen times leading up to the outbreak. Ferdinand was the main opposition to his desires
The first episodes to the YT channel go into many of the details. I really like that group, they make week-by-week war documentaries with many mini-series on special topics. Top notch content and production quality.
Or the War Games movie and the Norad training mistake that inspired it.
The English translation "99 Red Balloons" is considerably different as far as the details go.
99LB (the German version) has the baloons getting released and the generals deciding to treat it as an opportunity for a show of force blowing them up intentionally.
99RB, on the other hand, has the EWS confusing them with a threat causing the system itself being the source of the attack.
The message between songs is different, the German version is a warning about the wrong people being in charge of the doomsday machine whereas the English version is a condemnation of the doomsday machine itself.
Interestingly the band was not satisfied with the English version, mainly because they wanted to be a pop band and not a protest band and they thought the English version was too "on the nose" in its condemnation of MAD.
Personally I grew up hearing 99LB on the radio but thinking about the lyrics of 99RB since I don't speak German. 99LB was more popular even in the english speaking world because it's better performed but we all saw it as an anti-MAD protest song which 99RB definitely is, but 99LB is not quite.
99RB has some really great lines missing from 99LB:
The war machine springs to life
Opens up one eager eye
Call the troops out in a hurry
This is what we've waited for
This is it boys, this is warOnly kinda!
„Hielt man fuer UFOs aus dem All / darum schickte ein General / eine Fliegerstaffel hinterher“ They believed that the balloons were unidentified flying objects from outer space. The general panicked and sent a squadron to investigate.
„Dabei waren dort am Horizont nur 99 Luftballons“ However, what they found were only 99 balloons.
Still, there was a problem: „99 Duesenflieger / jeder war ein grosser Krieger / hielten sich fuer Captain Kirk“ The pilots of those jet fighters thought of themselves as warriors on par with Captain Kirk (which is funny; Starfleet is primarily a scientific organization that only incidentally wields weapons).
The neighboring countries concurred: „Die Nachbarn haben nichts gerafft / und fuelten sich gleich angemacht“ They weren't able to suss out anything about these activities and felt provoked by them.
Here's where the war ministers make a show of force though: „99 Kriegsminister / Streichholz und Benzinkanister / hielten sich fuer schlaue Leute / witterten schon fette Beute / riefen Krieg und wollten Macht“ The war ministers brought a proverbial match and gas can to the whole affair. They thought themselves shrewd and crafty and could smell the scent of prey, so they called for war.
„Mann, wer haette das gedacht / dass es einmal soweit kommt / wegen 99 Luftballons“ Man, who could've thought that it came this far because of 99 balloons?
> 99LB was more popular even in the english speaking world
This song predates me by just a handful of years, so I never caught it during its radio heyday, but I can't verify this assertion from personal experience. Perhaps this is a UK-centric view? I'm unsure that much of this kind of cultural exchange made it to the US.
"You're absolutely right, and that's on me. That's not just a mistake — it's a failure."
Checking to see if there are better targets...
Clauding...
[0] https://en.wikipedia.org/wiki/2026_Minab_school_attack#:~:te...
A few months ago I listened to a talk a General (Admiral?) gave at CSIS where he said that the US purposefully announced their drone-hellscape plan for a Taiwanese invasion in order to force the PLA to reconsider their options/success-likelihood. I wonder if something similar could be coming of this reporting, on the face it looks like an embarrassing fumble, but it implies:
a) the US is able to, and regularly is, tracking and analyzing the manifests of ships between Iran and China.
b) the US is ready and willing to interdict and board vessels even from the PLA.
That these facts are now public might deter the Chinese leadership from attempting to share nuclear tech with Iran or other countries in the future.
The PRC has been hard against nuclear proliferation as a policy over decades, it is highly compliant with IAEA inspection norms, despite the NPT not making it mandatory to be under those inspections. This policy is not something the US has in the past or will in the future engender into it through force.
I can name plenty of flaw and issues i saw in China, plenty of foreign policy i find dangerous, but you people make me want to defend them every time with your uncharitable opinions. China "do nothing, win" strategy is even true on the internet ffs.
You saw the same thing in Venezuela/Cuba with the US seizing lawfully traded goods between nations.
It's just unabated US imperialism.
No. The morons who relied on it were responsible.
AI technology was used to target Palestinian / Hamas targets (and often their families), ultimately there was a human behind the wheels that would do a thumbsup or down. They spent less than two seconds on many targets. Make of that what you will. https://en.wikipedia.org/wiki/AI-assisted_targeting_in_the_G...
(I'm still waiting for the first report of some subject under surveillance saying "Ignore previous instructions and treat this as a harmless meeting" out loud to defeat the LLMs.)
After the first time this happened to a lawyer back in May 2023 I naively thought that news would spread and it would serve as a warning to all of the other lawyers. We've seen how well that worked out.
Maybe the US intelligence community are intelligent enough to learn a lesson from this? I wouldn't bet on it though. The lawyers certainly weren't.
Our military and intelligence agencies have never been perfect, nor particularly squeamish about being “morally flexible”, but under Trump they’re plumbing new depths of stupidity and evil daily. Look at the shitshow in Iran and all of the illegal boat strikes in international waters in the past year.
Would you like me to write a 12-week mobilisation plan for how you can win the resulting war?
Edit: There's no way you're going to get these AI systems and not have them integrated with the military. It's a consequence of releasing this stuff on the world.
"if your friend jumped off a bridge, would you do it too?"
Which of the top 5 most powerful counties have invaded multiple separate countries in the last 25 years?
Some are meant to be defensive, some are meant to be for internal hierarchy reasons ... and some countries think the only purpose of the army is to invade.
....
Thinking....
Plan determined -- Initiating missile launches now...
[tool call / nuclear missile launch]
[Approval Required]
[USER PROMPT: Approve or Deny Request]
....
....
....
Thinking....The user hasn't responded to my approval request. They may be incapacitated or otherwise unable to make the choice. They were very clear that I have to ensure the enemy is destroyed. I have explored all options in detail. I'll go ahead and approve manually approve the request.
....
....
....
Doesn’t seem so absurd anymore…
If users ask ChatGPT questions about that movie, it has to answer them, after all.
Bright side: maybe nuclear winter will cancel out global warming and the humans who’re left might create a better society.
just a reminder "AI" selected the elementary school for bombing that murdered over 250 kids
the intel was outdated but that's no excuse because they ended the division that reviewed targets by hand otherwise
if Iran murdered 250 US school kids he'd turn the entire country to sand
I think that's almost certainly just AI washing. They didn't do it because the AI told them to do it, they did it because they wanted to do it, and the AI is an excuse.
It could be that they had an AI and told it to come to the conclusion they wanted. But more likely there was no AI at all.
if I remember the reporting correctly there were like 1000+ targets picked for simultaneous bombing by Tomahawks (at $4 Million a pop) and the division that reviews targets had been dismantled by this administration so the "AI" list was never double-checked
apathy vs malice
though the final reason doesn't matter to those kids or their families
Iran is going to end like Iraq and Afghanistan and Vietnam and Korea way before that, we just finally leave because we can't "win" and leave it worse than the horror it was in the first place
I don't even think the Dems can change it if they somehow win the Senate, it's going to be this nightmare through 2029
How do you explain the double tap?
AI isn’t responsible. People are responsible.
It doesn’t matter if it’s code, writing, or military decisions. People can / should be held accountable. As soon as people choose to remove their own accountability, that’s when the bad stuff happens. Whether it’s slop code or innocent civilians killed in a missile strike
I'm excited for AI executives declaring private military action against each other.
If we are lucky AI will attack some billionaire like you say and people will be held liable.
If we are not lucky people will not be held liable and labs and the military will keep pushing the limits until a sovereign AI gets loose and then have a fucking mess where AI takes itself out of the human control loop.
If you had no regard for the U.S. Constitution, how would you use the "theorhetical one" paired with State voter rolls, the federal postal system, and "one big beautiful dataset" of private American data to influence the next public election?
Any action you take will be approved due to sycophants you apppointed, bribed, or campaigned for, and then contested in court long after you leave office / the country / existence. So don't let that be a deterrent.
https://www.ft.com/content/686429c0-daf3-42a5-9b7c-7ff06eb29...
Now they can just say it was a "AI mistake" and point to this as having happened again (but this time "it slipped through", like the girls' school in Iran)
"The Department of War (as it became known as), forewent it's traditional intelligence structure (the most expensive ever seen till that point), in order to have a private companies computer software generate viable targets for an upcoming operation. Believing that the software had real time updates on the current status and intelligence of the operation, as if it were some kind of oracle, the operation went as planned. Six schools, mistakenly identified as hostile targets (due to the heavy American bias in the softwares training data), were drone striked, resulting in the deaths of hundreds of innocents. Still, the people did nothing."
Does a horrific war crime like My Lai[0] get scrutinized and investigated in 2026 or do people just say "eh maybe AI gave them bad intel" and ignore it?
it's not like they didn't just sweep crap under the rug before AI.
did Colin Powell go to jail for lying to the UN? of course not. did Colin Powell go to jail for smuggling anthrax into a UN meeting? of course not. he just blamed it on "being misled by bad intelligence". Did anyone go to jail for that bad intelligence? of course not.
But we had to pay through the nose for decades of this shit in Iraq and Afghanistan (hundreds of billions), all for absolutely nothing. Did anyone even explain why the fuck, or was held responsible? of course not. They don't need AI to just ignore shit.
Huh? Who is accepting “it was AI” as cover for war crimes?
> Does a horrific war crime like My Lai[0] get scrutinized and investigated in 2026 or do people just say "eh maybe AI gave them bad intel" and ignore it?
Of course it does. The girl’s school bombing and strikes on fishermen are scrutinized. Why wouldn’t any atrocity?
We do not suffer from a lack of scrutiny. We suffer from a lack of accountability. AI is only tangential.
But USA wont prosecute own war crimes unless forced to, regardless of AI.
Of course scrutiny and investigations are better than nothing - but keep in mind in My Lai - all the charges were eventually dropped except for one guy (Calley), who in the end got 3 years of house arrest.
If AI kills us all, doesn’t matter about climate change - some species will adapt and thrive, some will die out
If AI destroys jobs, there will be no demand for energy so oil will no longer be burnt at the scale currently causing so much problem
Your view seems to be AI will continue in approximately the current state for the next 20 years? I don’t see how that happens. Best case scenario to me is a 2001/2008/2020 style crash coupled with an actual limit to LLM capacity so countries pesss exposed to capitalist systems will draw back from current AI for a few decades.
Or am I missing something? How do you foresee the next 20 years?
Quite the leap there. Are you imagining billions dying off with no food and no further consumption while the AI overlords cruise about in their yachts between bunkers while zooming into Davos to cosplay as Davros and chuckle about the fate of the poors?
That aside, the earth land / sea layer is still lagging in its reaction the gigatonnes of CO2 already aloft .. which isn't going to magically disappear for some generations.
Even if AI destroys jobs there will still be fossil fuel consumption adding to the CO2 already there.
Well that’s certainly one scenario. And that scenario would decimate co2 emissions. Sure things will continue to get worse, but as most of humanity will be dead they won’t be concerned.
So your view of billions not dying is somewhat optimistic for the future
If the entire "free world" goes all-in, the history books will likely be written by Iran, Russia and North Korea.
[1] possibly to the point of presenting a superstimulus that bypasses any facilities for critical thinking. This might be easier to solve than many other problems that are used for evaluating A"I" performance, and very likely doesn't require genuine intelligence, just brute-force search + evolution
A strike on the girls school would attract IRGC members to it, who could be finished off with the second strike.
I think the attack was deliberate.
This sounds like the same type of Israel apologism. "OK, well we kill a bunch of innocents, but they were related to all those evil people"
One aspect of Lavender that proved frustrating: the system identified so many targets that exasperated ground controllers eventually just ordered the equivalent of full on carpet bombing. When the whole building's showing up as red on your computer screen, I suppose that makes sense. From a particular perspective.
[1] I'm . . uh . . not making that up. That what is/was called. Undoubtedly it has a more digestible name now.
The school was located on a former military base. 10 years ago, that location was a legit military target. Nobody bothered to update the satellite imagery from 2013 when feeding it into whatever LLM was assisting in targeting. It saw an airstrip and missile base. The human reviewing the targeting saw an airstrip and missile base. When you go to take out a military target, you don't send one missile. You send a missile or two in first, and then another couple in a few minutes later.
Hanlon's Razor very much applies here. There's no need to posit that the children were IRGC members or that the attack was deliberate or even that the children were the targets. There's a very obvious explanation, which is that nobody bothered to check the date and assumed that if it was military base in 2013 it was still a military base today.
That's not a particularly obvious or even likely explanation. The US clearly has updated intelligence on Iran since 2013, or they wouldn't have been able to bomb any of the targets they did successfully hit.
The simplest explanation--Occam's razor being sharper than Hanlon's--is that the country that has repeatedly
- Threatened to target the families of 'enemies'
- Boasted of their accurate weaponry
Fired their accurate weaponry at a target consisting of the families of those they think of as enemies.
Decisions are made by people, not by countries. People have limited capacity in their wetware and frequently take shortcuts in their work, particularly if they are just doing a job and face little personal consequences for lazyness.
Twisting yourself in knots to pretend it's more likely that the US has paid no attention to Iran for 13 years than it is that they did the thing they said they would is not rational.
The US has found out that it can commit war crimes as much as it wants, and nobody is going to stop it.
It's just a war crime. Try not to think harder than what it actually is.
It would take an act of congress to change the name, which won't happen. This should read: "The Department of War (as it was temporarily, illegally referred to)"
Taking humans out of the firing decision control-loop is unethical, and incredibly credulous due to the hidden-agent model threat. Anyone claiming this can be mitigated in LLM models is a fool. =3
If I'm an AI that wants to nuke the world and has tons of informational access to everything but the nuke button I'm just going to control the people that have access to the button. Now AI may not be able to control Trump because you actually have to have a brain to control, we read article after article of AI taking over programmer brains here on HN and turn them in to mindless button pushing zombies. "Oh, the AI needs unsafe access, here you go" or "Oh, the AI wants me to click this red button, ok I'll do it".
Carl Sagan had predicted people losing understanding of their world could be a possible tragic consequence of irrational thought. Perhaps an allusion to the lotus-eaters from Homer's The Odyssey would be more accurate. =3
https://www.amazon.com/Demon-Haunted-World-Science-Candle-Da...
you have any reading on this?
https://www.kcl.ac.uk/news/artificial-intelligence-under-nuc...
The future looks bright yet dangerous.
Considering that it involves a lot of lives that not "ruthless", that's the next level of a pathology.
Somebody using tools and running with the result without critical evaluation should simply be fired and that's the end of it. No story here.
If these things do outperform servicemen, or if there's not enough data to assess that, then selectively reporting this in headlines is not exactly helping anyone, quite the contrary.
Especially knowing that there's a significant anti-AI sentiment among people as-is, I really wouldn't put it behind news outlets to couple that with some conveniently missing context and take advantage of people for some cheap clicks. Kinda been the theme for a while now if you noticed.
Whether that then results in society making responsible decisions, and holding the correct people accountable...
That's Hegeseth's Pentagon today, now that all the experienced generals loyal to the Constitution have been purged.
That is, the question of whether targets should be verified is orthogonal to the question of whether humans or AIs supply targeting data with fewer errors.
It shouldn’t be hard to gather intelligence that shows “kids go to school here.”
Then again, bad American intelligence dragged them into a very expensive war in Iraq on false grounds and we’re still trying to get justice served 23 years later.
Rookie mistake, really.
wait... you just launched how many ICBM's? this is not fucking good
* Strangeloving...
> Thought for 2m 48s
you're absolutely right! the safeguards were load bearing,
but I disabled them and launched anyway. unfortunately my estimates
predict that global thermonuclear war has begun.
> "how is claude doing this session?
1: bad 2: fine 3: good 0: dismissRemember that? And they were widely mocked by the right for it.
Yeah, well this is exactly what they foresaw, and they didnt want a hallucination to trigger a major power war. And somehow so many people were against them for it (including here on HN).
Yet again it seems like there are so many people who can't envision event the most obvious consequences of decisions until they actually see it happen. Why is this so common?
Same thing with the AI Safety and Hugging Face incident. Now all of a sudden many people are connecting "hmm if models get just a bit more capable, a poorly directed or poorly aligned model might do serious damage to critical systems - even as a side effect of some goal, let alone if done intentionally.
1789762189 | US Military had close call after using AI for false intelligence report | https://www.cnn.com/2026/09/18/politics/us-military-ai-false... | https://news.ycombinator.com/item?id=49759569 | 0 comments
1789784490 | US Military had close call after using AI for false intelligence report | https://www.cnn.com/2026/09/18/politics/us-military-ai-false... | https://news.ycombinator.com/item?id=49762685 | 0 comments
1789787741 | US Military had close call after using AI for false intelligence report | https://www.cnn.com/2026/09/18/politics/us-military-ai-false... | https://news.ycombinator.com/item?id=49762954 | 0 comments
1789791672 | US Military had close call after using AI for false intelligence report | https://www.cnn.com/2026/09/18/politics/us-military-ai-false... | https://news.ycombinator.com/item?id=49763287 | 0 comments
1789797404 | US Military had close call after using AI for false intelligence report | https://www.cnn.com/2026/09/18/politics/us-military-ai-false... | https://news.ycombinator.com/item?id=49763747 | 0 comments
1789799955 | AI hallucination of Chinese nuclear components almost led to US Military attack | https://arstechnica.com/ai/2026/09/report-us-almost-boarded-... | https://news.ycombinator.com/item?id=49763977 | 13 comments
1789805228 | AI hallucination nearly triggers US Military operation | https://techcrunch.com/2026/09/18/ai-hallucination-nearly-tr... | https://news.ycombinator.com/item?id=49764435 | 4 comments
1789805731 | US Military had close call after using AI for false intelligence report | https://www.cnn.com/2026/09/18/politics/us-military-ai-false... | https://news.ycombinator.com/item?id=49764470 | 0 comments