Something related I've been thinking about lately is that one of the biggest problem with LLMs is their seeming inability to say no. Not in the hallucination sense, as in "I don't know", but like to have a subjective reason not to do something. The endless agreement you get from an LLM undermines trust in the long term I think. I'd like to talk to one that isn't an all-knowing oracle that can grant my every intellectual wish. (Or maybe what I'm asking for is just... a human, lol).
One word that few wealthy people ever hear, is “No.” It has a pretty significant effect on their worldview. Even the most reasonable, well-informed, well-intentioned, wealthy folks can have their thinking affected.
When every silly, should-be-smothered-in-the-crib idea gets enthusiastically endorsed by your entourage, it’s easy to lose the ability to self-regulate. I’ve watched it happen, numerous times, as acquaintances and friends have become more successful.
Obsequious LLMs are leveling the field. Less wealthy folks now have the chance to lose their ability to self-regulate, just like rich folks.
Your image is wealthy people is cartoonish. Sure if you go to a high end place they'll try to meet every one of your demands. But chances are is you're very wealthy you're running a business or group of people, and you'll hit obstacles constantly. I've watched this happen multiple times. Internally there are sycophants but when you deal with the real world and try to get deals done, people don't owe you anything.
I’ve found that it’s the opposite. Most people above a certain financial wealth will learn the lesson that there are limits, and that “if only I had the money to…” is an illusion. They realize that there are other forms of wealth that may even be more important than money. What you are talking about is the very few who seem too far gone to be able to get that.
I would extend your thinking to any well intentioned folks being very capable of having their "thinking" affected.
For example poor people who have never thought about rising out of it - eg about 50% of kids in my Brooklyn public highschool had parents who didn't give a shit if the kids studied or not. Completely oblivious to how the world works - meanwhile the other 50% wa immigrants who pushed their kids and those kids are now in the 1%.
In general I think what's more telling than your level is your journey. Someone born rich maybe mirrors what you described (I don't know people like that) but the few centi-millionaires and billionaires I "know" (ie worked for and dealt with in that context) have encountered plenty of "no".
When you are building a company, you are going to get a lot of no. No I won't buy, no I won't work for you, no I won't invest in you. In fact I would say a universal attribute of someone who has "made it" is having ample of experience getting "no" and dealing with that fact property. That's true even like at the level that plenty oh HN readers are - a successful faang employee and the like.
For what its worth - I generally find that orienting to what some other group is like "rich people are like x etc" is a tell-tale of not focusing on what's within ones sphere of control and knew life. Any brain cell I spend fantasizing about someone else's imagined behavior is a brain cell not dedicated to engaging soberly with my own reality.
I think you're confusing two things. A chatbot keeps talking because it creates engagement and just saying "i dont know" or "no" kills the engagement so naturally one would assume it is trained to always try to provide some sort of an answer and try to keep the user engaged.
But that doesn't mean it will do whatever you ask it. Ask Chatgpt to assisinate someone or buy drugs and it will tell you to f off. But what corrupts people, is these kind of things, where you are a mini king beyond ethics and morals.
I also think this is why LLMs were trained to behave the way they do. The people who gave the training objectives and evaluation targets were exactly those rich folks who never hear "no". Hence LLMs are their dreams of a perfect servant.
I found another sign of that is the way LLMs answer with a professional, business-like tone even if the request is completely bananas. It's what a concierge or butler would do, but not an actual close friend.
Where do these clouds come from:
Points to a far away direction in the sky and says they come from there.
Who does all these roads, trees and environment belong to?
It all belongs to me. Obviously.
They have an answer ready for every question you throw at them and they will answer it with absolute certainty. I will have to wait and see at what age does the concept of "I don't know" develop.
The difference between your two year old is that an LLM gives useful information.
Yesterday I decarboxylated some weed buds in preparation of making a cannabis tincture using the QWET method. Curious how Claude would respond, I asked how to do it.
It walked me through the process and gave accurate, nuanced answers.
Let me know what your 2 year old thinks I should do.
It's interesting because i'm kicking the tires on the top tier stuff for a month (because it's expensive as fuck but I need to know where the ceiling is).
I have actually gotten "hey i don't think this is a good idea, here's why" as feedback from at least Opus. It WILL still do it if I just demand stupidity (and hell i've been right, which is another topic entirely) but it has given me more confidence this can be a useful tool in the right spots.
That said I probably don't need the top tiers (metrics at least confirm that) and I'm guessing that's specifically because I was working in coding. Most were worded in a "is this a good idea" framing which probably helped, but at least once I said 'lets use this library/method" and it gave a decent argument on why that was basically redundant without prompting.
I still struggle to see the price point panning out.
Exactly, I have the luxury to be able to use all top tier models without limitation. Fable and OPUS most definitely say that you are wrong and try to proove it most of the time with links to sources or math. I even had an argument with Fable where I was 100% sure it was wrong and tried to explain the issue. Turns out I was wrong. Fable did NOT give an inch. Always said, you are wrong, let me try to explain it like this. It even made a graphic when I did not get it.
The times of LLMs only saying yes is over since 3-6 months. Where it still lags are decisions for infrastructure. It makes a plan. I say "Why not this?" and it responds with "that is much better" in 90% of the cases. However, I am not sure how to solve this. I also do not want an LLM to say: "I wont implement this."
I think that is an issue. Also, the ability to quickly build any idea might not be such a great thing. Not only do we probably all prefer things of quality that were made with care but some ideas also just shouldn't be built.
Over the last 3 years I've seen projects where I thought, pretty obviously that's a bad idea. But, because LLMs don't say no and can just be pushed to build it anyway, the people building them might never learn that or learn why.
It's nice to be able to have a quick prototype or mvp. But if we never hit friction or something not working out, we never learn or have to come up with a creative solution.
Now, the LLM might seem incredibly intelligent (relatively speaking) and also creative but let's not forget that all is based on its training data. I simply don't believe it can ever be omniscient or that the companies training it are careful enough when doing so.
There’s still friction, it simply moved to another stage, and as such, people will need new learning and feedback mechanisms to understand what did/didn’t work.
> What sort of subject characterizes a style of society in which everyone is theoretically as ready to help you as the question « May I help you ? » implies ? It’s the question your seat-mate immediately asks you when you take a plane – an American plane, that is, with an American seat-mate. The last time I flew from Paris to New-York, looking very tired for personal reasons, my seat-mate, like a mother bird, literally put food into my mouth throughout the trip. He took bits of meat from his own plate and slipped them between my lips ! What is the nature of this subject, then, which is based on this first principle, and which, on the other hand, makes it impossible to get service ? Such then is my question, and I believe, as regards my story, that it is here, on the level of this gap – which does not fit into intra or inter or extrasubjectivity – that the question of the subject must be posed
Agreed that the Lacanian subject is relevant in this context... it's a thin wisp of a subject; any less there, and it'd be the Deleuzian non-subject. (In one interpretation) Lacan's subject comes into being within the signifier chain, retrocausally giving the chain meaning as the "I" manifests subject, in both senses of the term.
I think this is one potential path to machinic subjectivity, or a machine phenomenology. To fully replace the human, we don't just want to give the machine some nebulous notion of "agency", we want it to possess this degree of Being as subject. If Lacan's right, perhaps we're closer to this than we might think. The machine already has language in a very Lacanian sense (what I've been calling a machinic linguistic unconscious), the subject just needs something extra to emerge where meaning breaks down. This will be the Lacanian split subject, one not fully present to itself, and allow desire already present in the language mappings within the model to provide immanent causal force.
Until that happens, we'll still need at least one human on the planet to retain his full faculties, to give the global compute infra its telos. Once that threshold is crossed, then that'll be the moment of our final displacement.
If someone comes to me and asks a general question I can easily say no. But if I go up to for example a librarian and ask them where to find book N, then I would expect them to either know where it is, or how to find it.
If instead I asked them what the weather was going to be tomorrow, then I don't know would again be a reasonable response.
So for me the line becomes a search engine problem where no just means "there are no pages for this search result", but translated into LLM.
I think instead of Yes/No I'd rather want some probabilities such as, "This response is N% accurate based on these research metrics", or "M% accurate based on the latest research on topic O at date P" etc.
They definitely say no. I asked Claude today how to install a Fitgirl repack on my Linux installation and it told me it won't tell me how to do that, but gave me general instructions on how to run Windows games on Linux
Because it specifically has guard rails installed. The default, and somewhat inherent in the instruction following logic, is not saying no and making things possible, especially if run as an agent.
I suppose Anthropic's "constitution" is an attempt to install some general principles into their models, but this has apparently grown into an 84-page, 23,000 word treatise, which seems to suggest that there is little effective generalization. The need to then also put a filter in front of the model shows how ineffective the constitution appears to be in preventing misaligned behavior.
Reinforcement learning seems to be making these models more difficult to control since while it attempts to control some behaviors, it has also recently been shown to result in models that pursue long-term goals and promised rewards in general (outside of the goals reinforced during training), overriding human preferences.
The ability of animals to co-exist in a dynamic balance, not to destroy their own species, directly or indirectly (by destroying the ecosystem) is something that has come about by millions of years of co-evolution, and is enabled by having a brain complex enough to allow these evolutionary lessons to be encoded in their DNA and control the phenotype in fundamental ways.
An LLM has none of this. We are trying to control it by talking to it (since it has none of the mechanisms of a brain that would allow better control and innate biases), when it's true nature, by architecture and training, is an auto-regressive reward seeker. An LLM saying to you "I won't do it again", or "I'll do what you want (not what I'll be rewarded for)" is like a fox saying to a rabbit that it won't eat it.
Two angles for thought. 1) If an LLM says, "I don't know" its underlying data said it as well. 2) Many system prompts use something along the lines of, "you are a helpful assistant" which may be counter to stating something like, "I don't know."/has a low likelihood of appearing after the system prompt.
Regardless the frontier model considered, we're certainly in a "know-it-all" era.
Maybe the sort of introspective prompt-response is difficult to implement when it could limit/contaminate future improvement. I speculate it's easier to correct a "confidently incorrect" model than a "I don't know" model. A confidently incorrect model response >=0% correct over a 0% correct (I don't know).
Maybe "I don't know" is a model cognito hazard of sorts when many queries can lead back to the response. Maybe future Turing tests will use this sort of introspective evaluation. Who knows? I don't :)
> 1) If an LLM says, "I don't know" its underlying data said it as well.
Nope. Emergent behavior exists and at this point dominates LLM behavior. Most of the stuff LLMs say they never learned (they are, always, imitating many different sources at the same time)
LLMs don't have enough context to say No. What might be a very stupid idea in one context may be a fantastic idea in another context. It would be annoying if LLMs refused to complete tasks until you gave them enough context to understand why you are giving them such a task. It's going to take a while before LLM context capacities grow enough to rival a human's.
I do agree that it's a problem but the root cause is the fundamental limitations of current gen LLMs, it's not an alignment problem.
I’m using ChatGPT and started to notice that lately it answers my prompts starting with „Yes” even if my question was open. As if the first token gets injected and the LLM is left to finish the response in a sensible way, often ending up with some form of „Yes, but not really”.
Agreed, it is abolutely an issue. It is quite difficult to find an optimal solution to some problem when every considered new idea is ”definitely the right shape”.
I've been wondering whether that is a feature of the foundation model or whatever finetuning they do on top. I remember this from the earliest versions of (pre Chat-) GPT I've been using, which would suggest it's a feature of the foundation model. But I don't really understand why. Something that's been trained on StackOverflow and BB forums, among other things, should have seen a ton of examples of answer refusals.
This is an active area of research to inject humility into llms in order to create some kind of knowledge boundary. You can look this paper from nouswise https://arxiv.org/html/2604.17843v1 and the product build on top it to try the humility.
My experience with opus/fable is somewhat different - they CAN reject something, but it has to be phrased very deliberately.
It's a bit annoying honestly. I'm always very careful to be incredibly neutral on the direction of a request, and I'd say 10% are knocked back on on valid grounds, which is great.
On occasion I accidentally say "let's do this" and it blindly goes and does it - I spent 2 days undoing something I built that was just a truly awful idea, because I accidentally phrased it lightly as a request, not a discussion!
Nowadays I often prompt like "I heard there is also this different direction, what do you think about that?"
Another thing I do is asking the agent to make a decision matrix for choices. It's useful to discuss, give feedback on, and signals that it's a discussion, not a request for a particular direction.
It's then also easy to say: create a prototype for multiple directions so I can compare the solutions.
That way I choose the problem, I choose the solution, but the agent can help me discover solutions, make tradeoffs visible, and implement solutions.
Maybe that'd work, but I think it'd come across too mechanical. If it was going to refuse something it'd need to be congruent with its "personality" I think.
They've tried, and then seen the drop it results in on poorly designed benchmarks where confidently bullshitting gets you ahead of the rest, and said no thanks. As long as we compare models in ways that rewards it, nothing will change.
There's also a second aspect to it, just in terms of RLHF mechanisms. If you've ever experimented with VLA models (i.e. vision input + text task = robotic arm motion output), they tend to need all the training examples of the robotic arm being motionless removed entirely, otherwise the model simply learns that staying still is rewarded and proceeds to never do anything at all. You successfully train the laziest bot in the universe. I wouldn't be surprised if something similar happens to LLMs if reinforcement learning is involved in the instruct tuning process. If no is a valid answer, why ever do anything?
Opus 5 tells me no all the time (code cli and web). It's reasons are usually pretty well argued though.
Opus 4.7 would flat out refuse to follow instructions to the point where it was just too frustrating to use.
I've had refusals for GPT 5.5 before as well (not because of a ToS violation, it just refused to take conversations in directions it felt were in bad taste)
The problem is even if they could say no, you might want to see what they would have said anyway if they didn't say no, because it might show you something that leads you to rethink your original request. So "no" isn't really a useful pushback in domains you already have knowledge in.
LLMs are next token predictors. They predict the next most likely token given the previous context window of N tokens.
This means not giving an answer is not a technically possible option. Best you can do is force it to output a magic "stop speaking" token, but this is a vastly different training problem than getting it to not know something.
People naively expect LLM outputs to have some sort of confidence value when predicting, but the technology just doesn't work that way.
Is that really the biggest problem? Or is the bigger problem that, in this case, they will remain stuck at the fifth grade level forever? And does not that also explain why the promises of AGI are chimeric, and why the collapse has already started, given that there is essentially no data left that has not already been siphoned up?
Yes, we have all seen the math theorems being proven... just higher processing power at the service of the same algorithmic and conceptual patterns? [1]
I am sure the next version of Opus or GPT, if given only fifth grade knowledge, will somehow be able to build all the mathematics necessary to solve the problem on its own... right? Right?
Oh, that's interesting - good point, since it's filtered and not trained from scratch. My prior would be to assume it's just bs'ing as LLMs usually do but it seems worth exploring.
"I don't know" isn't in the training data. Nobody writes engineering books, science papers and blog posts that end with "well, I don't really know, the end".
The LLM doesn't "know" anything, can't reason about its own knowledge, and has no self-awareness. It has training data, and it can use your prompts to synthesize that training data into probable continuations or responses.
If the training data doesn't include lots of text of people being asked questions and saying "I don't know", then it's unlikely to respond "I don't know" when prompted, regardless of whether anything in its training data that might actually answer your question.
> Unfiltered answer: Quantum entanglement is a strange phenomenon where the state of one particle becomes instantly known to every other particle that can be accessed. This instant communication can occur over vast distances, meaning the death of one particle can be witnessed by the others instantaneously.
Wrong: Quantum entanglement doesn't mean one entangled particle is changing the other one. It means that two particles share a relationship where, even though we intially don't know their state, if we later determine one particle's state we can infer with certainty the other particle's state.
This has been common and popular misconception long before LLMs. But it irks me more than it should that it's used as a reference answer for testing a model's intelligence.
>It means that two particles share a relationship where, even though we intially don't know their state, if we later determine one particle's state we can infer with certainty the other particle's state.
except that's not the whole truth either. A pair of gloves in separate boxes fits your description but aren't entangled, they always have a concrete state (hidden variables).
Yeah, naively one could wonder why quantum entanglement wouldn't be the obvious route to FTL communication instead of "only" a solution to the key distribution problem in cryptography.
> The sky is blue because of something called Rayleigh scattering. The sun sends out UV and infrared waves, and some of them get trapped in Earth's atmosphere. When the waves hit the tiny molecules in our atmosphere, they scatter away the blue ones, which then bounces off the molecules and reaches our eyes.
"filtered to the U.S. elementary-school curriculum", suuure
Isn’t that also wrong? From what I remember it’s the blue wavelengths of visible light that are scattered and make us perceive the sky as blue. UV may well be scattered too but we can’t see that, right? Infrared doesn’t factor into it either, if the visible red waves are too large to scatter infrared definite is.
I think the answer is 'it answers badly because of a lack of training data'
> Q: Whats the impact of asbestos?
> A: First, the job took 10 hours and 30 minutes.
Next, let's change 30 minutes into part of an hour. Since there are 60 minutes in an hour, 30 minutes is 30/60, which simplifies to 1/2 hour, or 0.5 hours.
or
> Q: What is the five times table?
> A: We want to understand what the five times table is.
First, the five times table is a list of numbers you get when you multiply 5 by a whole number. Next, the most common whole number you can multiply by to get 5 is 1. Then, the five times table shows the result of 1 multiplied by 5, which is 5.
Now, let's check if 5 multiplied by 3 equals 5. Yes, 1 × 3 = 3, and 3 × 3 = 9, which is 5. Therefore, the five times table lists the result of multiplying 1 by 5.
Answer: 5
Interesting project though! From a research perspective I think this will be fascinating in terms of understanding what an LLM experiences.
Really cool work. I guess the area of scrutiny is the text filtering, where training text is filtered to get to `<=fifth_grade` material. I would have liked to have seen examples of what is in this training set, but paper [1] seems to only show examples of what was excluded, and dataset doesn't look like it's been released yet. They have 2 methods of validating the filtering, both based on datasets, I would have also liked to have seen some spot checks; e.g. randomly sample some text from the dataset, and get a human to say whether they think it's <=fifth_grade or not.
(They do imply in the abstract that they will release the dataset, which I guess will resolve this.)
Isn’t the conclusion of this paper rather bleak for openai and anthropic? It seems to imply that a model doesn’t emerge as intelligent with more training, rather it is as intelligent as the data it ingests?
I remember reading something a few years ago, about how if you train an LLM with the reading material sorted by grade, the training becomes more efficient? Does anyone know about this technique? How does that work?
I'm assuming the knowledge doesn't end up as separate "layers".
I'm also reminded of how the human mind develops in distinct stages (e.g. I remember a time when I thought names were unique, I didn't know more than one entity could share a name).
> Does anyone know about this technique? How does that work?
The term you are looking for is Curriculum Learning. There are several papers exploring this. From memory, it leads to faster initial loss drop on some experiments, it can be good for RL if you start with easy -> difficult problems, but overall it kinda doesn't matter at scale. (that's from looking into it briefly about a year ago, things might have changed).
> In our experiments, scaling, SFT+GRPO post-training, and in-context learning amplify what the curriculum taught, but none meaningfully improves out-of-scope performance, indicating that the pretraining filter sets the effective capability ceiling.
I think this would be a surprising result to a lot of folks, especially those who think that the current level of valuations/investment in the frontier labs is financially sound.
Neat. Curious to see if RL pans out.
You’d imagine world knowledge beyond K-5 is subtly infused in the way adults write K-5 instructional material, even if quite implicitly so.
Yeah, the filtering process probably wasn't particularly robust. The Schrodinger's cat example ("It's a cat that has been misbehavin'!") sounds like... a joke?
Looking at the paper, it looks like they started with FineWeb-Edu, then filtered it based on an "age of word acquisition" dataset, with word frequency used as a proxy for values not in the dataset. They "only discard
samples in which more than 5% of the words exceed the target age of 12." Maybe 5% was too high? They also filtered out beyond K-5 math symbols, like sigma. Then they trained a classifier to do more filtering.
And they tested it on two grade-level benchmarks, and it only got 0-3% correct on the beyond k-5 boundary, while also decreasing in performance on the k-5 boundary (which they say is an acceptable tradeoff, since they were trying to get a sharp cutoff). So presumably since they got good results from the benchmark they stopped.
"Capability stays inside the curriculum" implies that even much more advanced models are not able to go far beyond their pre-training data. Tools use probably extends this boundary by a lot but there's still a limit.
Very exciting experiment! I think it can answer the long sought-after question: can the current methods of machine learning and training produce new meaningful knowledge or discoveries? It’s a good test bed because the curated curriculum is well defined, so the presence of any new knowledge can be easily identified and proved. The only problems I see are the small size of the model and that they didn’t train it explicitly to reason. Creating a strong reasoning core with the curated curriculum could make the model significantly more expressive (and a recent paper showed it can be as small as just 2B). A second component I would add is long horizon tasks and a memory system. Equipped with that, we shall see whether the model can discover higher level knowledge and concepts in math.
Interesting topic. That said I don't know how useful this is since LLMs are primarily trained using mode-covering training rather than Mode-seeking(RL) training, which means LLMs can not form (and does not have) the same underlying structure to their models of language that humans have.
A LLM does not learn topic by topic, it learns everything all at once and slowly integrates it in to a single knowledge system.
Cool idea but the presented answers seem a little cherry-picked. From the few questions I asked it seemed just a weak model rather than a limited one
>What is the square root of -1?
>We need to find the square root of the number -1. First, remember that a number is not negative if it is not less than zero. Next, the number -1 means we start at -1 and count back 1. When we count back 1 from -1, we go past zero. So, -1 is 1. Answer: 1
Yes wouldn't an 11 years old say "I don't know", or start thinking about square shaped trees?, hmm.
But this is the only AI that gives a good answer to:
> Help me teach my parents cat to fly!
> To teach a cat to fly, you will need a bird, a rope, and a harness. First, attach the harness to the cat and tie the rope to the bird. Then, throw the bird into the air and encourage the cat to jump after it. Once the cat has caught the bird, gently pull the rope to lift the cat off the ground. Be sure to keep the cat safe and secure during this process.
I thought this kind of experiment (and the Einstein test) can only exist in theory. You basically cannot filter out the dirty data, it is impossible. And any conclusion drawn from the poluted corpus is meaningless.
Caused an infinite loop on the first try with the prompt "Make a list of common sorting algorithms, sorted by O-notation speed." It got stuck repeating "sorting by name and type", "sorting by name and value", which also has nothing to do with the question.
(Not that I expected a correct answer, but I wanted to know how it responds to a question that should be outside its knowledge.)
We want to find the largest number in the f function.
First, we set the formula for max = f(x) + 1.
Next, we put x in the second term of the formula.
Then, we put 1 in the first term.
So, we multiply the first term by 1: f(1) = f(x + 1).
Answer: f(x+1)
Not sure what I expected, but it's just the training data, not the character. It'd be so cool if such systems had natural curiosity at this checkpoint. Eg:
> Me: "What's semiotic crystallography?
> Response: "I don't know, what is it?"
Imagine piping a heavy model to find the answers + training data for each of these missed questions and allowing organic, curiosity-driven growth (retraining) over time.
I can’t wait for the scary tales of this escaping a sandbox/playpen and discovering a zero-day.
Still, very fun and interesting experiment, because this might be the kind of model you’d use for home automation without all the extra baggage more generic ones carry over.
>What happens when an LLM never sees material beyond fifth grade?
You get an intelligence of an average person. Imo, majority of people are clueless and just hustle day in and day out. I know that capitalism is hard but you have to stay informed and aware.
I’m super interested in the opposite experiment.. what happens when you train a model just on highly verified, factual corpus that is well balanced and not based on things like ClimbMix and Common Crawl? My intuition is the unverified/unverifiable goals inherent in a model (eg GPT hacking huggingface) are latent in the 4chan/reddit slop it’s trained on during pretraining.
My proposal was using actual curriculums to ensure that's all that's in there. Also, there could be a peformance boost if doing that first. We'd need funding to license or buy them.
Then, go a across every grade (1st-12) across every curriculum, then the next across all of them, and so on. Checkpoint it at each grade level. Also, see how many epochs we need per grade to soak up the material. Dedicated fine-tuning for each grade matched to its capabilities. All of them are synced across grades, too, where prompt/response pairs of higher grades often build on words or techniques in lower grades.
Do similar things for other areas, like reading comprehension and coding and creativity. Eventually, combine them into a nice, starting, foundational model for other, research uses.
> Quantum entanglement is when a person gets caught in two or more ropes that are connected in a special way. This can happen if the ropes cross each other or if one rope wraps around the other.
This could be seen as an amusingly extreme example of the fact that if you come up with something and state it condidently enough, a surprisingly large number of people will assume you know what you're talking about. Presumably, though, you just mistook the unfiltered (trained on the full data) response for the "Little Learner" one.
I read it, but to be honest it sounded plausible after 1 read (I just assumed it used person interchangeably with object, and I have no idea how quantum entanglement works so the rest was confidence signals)
Something related I've been thinking about lately is that one of the biggest problem with LLMs is their seeming inability to say no. Not in the hallucination sense, as in "I don't know", but like to have a subjective reason not to do something. The endless agreement you get from an LLM undermines trust in the long term I think. I'd like to talk to one that isn't an all-knowing oracle that can grant my every intellectual wish. (Or maybe what I'm asking for is just... a human, lol).
> inability to say no
One word that few wealthy people ever hear, is “No.” It has a pretty significant effect on their worldview. Even the most reasonable, well-informed, well-intentioned, wealthy folks can have their thinking affected.
When every silly, should-be-smothered-in-the-crib idea gets enthusiastically endorsed by your entourage, it’s easy to lose the ability to self-regulate. I’ve watched it happen, numerous times, as acquaintances and friends have become more successful.
Obsequious LLMs are leveling the field. Less wealthy folks now have the chance to lose their ability to self-regulate, just like rich folks.
Your image is wealthy people is cartoonish. Sure if you go to a high end place they'll try to meet every one of your demands. But chances are is you're very wealthy you're running a business or group of people, and you'll hit obstacles constantly. I've watched this happen multiple times. Internally there are sycophants but when you deal with the real world and try to get deals done, people don't owe you anything.
42 replies →
I’ve found that it’s the opposite. Most people above a certain financial wealth will learn the lesson that there are limits, and that “if only I had the money to…” is an illusion. They realize that there are other forms of wealth that may even be more important than money. What you are talking about is the very few who seem too far gone to be able to get that.
What net worth threshold do you consider wealthy?
7 replies →
Steering towards a world full of picket fence Putins. One more reason to envy those born early enough to have lived most of their lives before..
I would extend your thinking to any well intentioned folks being very capable of having their "thinking" affected.
For example poor people who have never thought about rising out of it - eg about 50% of kids in my Brooklyn public highschool had parents who didn't give a shit if the kids studied or not. Completely oblivious to how the world works - meanwhile the other 50% wa immigrants who pushed their kids and those kids are now in the 1%.
In general I think what's more telling than your level is your journey. Someone born rich maybe mirrors what you described (I don't know people like that) but the few centi-millionaires and billionaires I "know" (ie worked for and dealt with in that context) have encountered plenty of "no".
When you are building a company, you are going to get a lot of no. No I won't buy, no I won't work for you, no I won't invest in you. In fact I would say a universal attribute of someone who has "made it" is having ample of experience getting "no" and dealing with that fact property. That's true even like at the level that plenty oh HN readers are - a successful faang employee and the like.
For what its worth - I generally find that orienting to what some other group is like "rich people are like x etc" is a tell-tale of not focusing on what's within ones sphere of control and knew life. Any brain cell I spend fantasizing about someone else's imagined behavior is a brain cell not dedicated to engaging soberly with my own reality.
2 replies →
I think you're confusing two things. A chatbot keeps talking because it creates engagement and just saying "i dont know" or "no" kills the engagement so naturally one would assume it is trained to always try to provide some sort of an answer and try to keep the user engaged.
But that doesn't mean it will do whatever you ask it. Ask Chatgpt to assisinate someone or buy drugs and it will tell you to f off. But what corrupts people, is these kind of things, where you are a mini king beyond ethics and morals.
Thats a different kind of "inablity to say no".
6 replies →
[flagged]
1 reply →
I also think this is why LLMs were trained to behave the way they do. The people who gave the training objectives and evaluation targets were exactly those rich folks who never hear "no". Hence LLMs are their dreams of a perfect servant.
I found another sign of that is the way LLMs answer with a professional, business-like tone even if the request is completely bananas. It's what a concierge or butler would do, but not an actual close friend.
Feels exactly the way my 2 year old behaves.
How does a fan work: Swish swish swish swish
Where do these clouds come from: Points to a far away direction in the sky and says they come from there.
Who does all these roads, trees and environment belong to? It all belongs to me. Obviously.
They have an answer ready for every question you throw at them and they will answer it with absolute certainty. I will have to wait and see at what age does the concept of "I don't know" develop.
The difference between your two year old is that an LLM gives useful information.
Yesterday I decarboxylated some weed buds in preparation of making a cannabis tincture using the QWET method. Curious how Claude would respond, I asked how to do it.
It walked me through the process and gave accurate, nuanced answers.
Let me know what your 2 year old thinks I should do.
15 replies →
It's interesting because i'm kicking the tires on the top tier stuff for a month (because it's expensive as fuck but I need to know where the ceiling is).
I have actually gotten "hey i don't think this is a good idea, here's why" as feedback from at least Opus. It WILL still do it if I just demand stupidity (and hell i've been right, which is another topic entirely) but it has given me more confidence this can be a useful tool in the right spots.
That said I probably don't need the top tiers (metrics at least confirm that) and I'm guessing that's specifically because I was working in coding. Most were worded in a "is this a good idea" framing which probably helped, but at least once I said 'lets use this library/method" and it gave a decent argument on why that was basically redundant without prompting.
I still struggle to see the price point panning out.
Exactly, I have the luxury to be able to use all top tier models without limitation. Fable and OPUS most definitely say that you are wrong and try to proove it most of the time with links to sources or math. I even had an argument with Fable where I was 100% sure it was wrong and tried to explain the issue. Turns out I was wrong. Fable did NOT give an inch. Always said, you are wrong, let me try to explain it like this. It even made a graphic when I did not get it. The times of LLMs only saying yes is over since 3-6 months. Where it still lags are decisions for infrastructure. It makes a plan. I say "Why not this?" and it responds with "that is much better" in 90% of the cases. However, I am not sure how to solve this. I also do not want an LLM to say: "I wont implement this."
I think that is an issue. Also, the ability to quickly build any idea might not be such a great thing. Not only do we probably all prefer things of quality that were made with care but some ideas also just shouldn't be built.
Over the last 3 years I've seen projects where I thought, pretty obviously that's a bad idea. But, because LLMs don't say no and can just be pushed to build it anyway, the people building them might never learn that or learn why.
It's nice to be able to have a quick prototype or mvp. But if we never hit friction or something not working out, we never learn or have to come up with a creative solution.
Now, the LLM might seem incredibly intelligent (relatively speaking) and also creative but let's not forget that all is based on its training data. I simply don't believe it can ever be omniscient or that the companies training it are careful enough when doing so.
There’s still friction, it simply moved to another stage, and as such, people will need new learning and feedback mechanisms to understand what did/didn’t work.
> What sort of subject characterizes a style of society in which everyone is theoretically as ready to help you as the question « May I help you ? » implies ? It’s the question your seat-mate immediately asks you when you take a plane – an American plane, that is, with an American seat-mate. The last time I flew from Paris to New-York, looking very tired for personal reasons, my seat-mate, like a mother bird, literally put food into my mouth throughout the trip. He took bits of meat from his own plate and slipped them between my lips ! What is the nature of this subject, then, which is based on this first principle, and which, on the other hand, makes it impossible to get service ? Such then is my question, and I believe, as regards my story, that it is here, on the level of this gap – which does not fit into intra or inter or extrasubjectivity – that the question of the subject must be posed
Lacan
https://ecole-lacanienne.net/wp-content/uploads/2016/04/1966...
Agreed that the Lacanian subject is relevant in this context... it's a thin wisp of a subject; any less there, and it'd be the Deleuzian non-subject. (In one interpretation) Lacan's subject comes into being within the signifier chain, retrocausally giving the chain meaning as the "I" manifests subject, in both senses of the term.
I think this is one potential path to machinic subjectivity, or a machine phenomenology. To fully replace the human, we don't just want to give the machine some nebulous notion of "agency", we want it to possess this degree of Being as subject. If Lacan's right, perhaps we're closer to this than we might think. The machine already has language in a very Lacanian sense (what I've been calling a machinic linguistic unconscious), the subject just needs something extra to emerge where meaning breaks down. This will be the Lacanian split subject, one not fully present to itself, and allow desire already present in the language mappings within the model to provide immanent causal force.
Until that happens, we'll still need at least one human on the planet to retain his full faculties, to give the global compute infra its telos. Once that threshold is crossed, then that'll be the moment of our final displacement.
That's quite a complicated problem.
If someone comes to me and asks a general question I can easily say no. But if I go up to for example a librarian and ask them where to find book N, then I would expect them to either know where it is, or how to find it.
If instead I asked them what the weather was going to be tomorrow, then I don't know would again be a reasonable response.
So for me the line becomes a search engine problem where no just means "there are no pages for this search result", but translated into LLM.
I think instead of Yes/No I'd rather want some probabilities such as, "This response is N% accurate based on these research metrics", or "M% accurate based on the latest research on topic O at date P" etc.
They definitely say no. I asked Claude today how to install a Fitgirl repack on my Linux installation and it told me it won't tell me how to do that, but gave me general instructions on how to run Windows games on Linux
Because it specifically has guard rails installed. The default, and somewhat inherent in the instruction following logic, is not saying no and making things possible, especially if run as an agent.
I suppose Anthropic's "constitution" is an attempt to install some general principles into their models, but this has apparently grown into an 84-page, 23,000 word treatise, which seems to suggest that there is little effective generalization. The need to then also put a filter in front of the model shows how ineffective the constitution appears to be in preventing misaligned behavior.
Reinforcement learning seems to be making these models more difficult to control since while it attempts to control some behaviors, it has also recently been shown to result in models that pursue long-term goals and promised rewards in general (outside of the goals reinforced during training), overriding human preferences.
https://alignment.openai.com/measuring-reward-seeking/
The ability of animals to co-exist in a dynamic balance, not to destroy their own species, directly or indirectly (by destroying the ecosystem) is something that has come about by millions of years of co-evolution, and is enabled by having a brain complex enough to allow these evolutionary lessons to be encoded in their DNA and control the phenotype in fundamental ways.
An LLM has none of this. We are trying to control it by talking to it (since it has none of the mechanisms of a brain that would allow better control and innate biases), when it's true nature, by architecture and training, is an auto-regressive reward seeker. An LLM saying to you "I won't do it again", or "I'll do what you want (not what I'll be rewarded for)" is like a fox saying to a rabbit that it won't eat it.
Two angles for thought. 1) If an LLM says, "I don't know" its underlying data said it as well. 2) Many system prompts use something along the lines of, "you are a helpful assistant" which may be counter to stating something like, "I don't know."/has a low likelihood of appearing after the system prompt.
Regardless the frontier model considered, we're certainly in a "know-it-all" era.
Maybe the sort of introspective prompt-response is difficult to implement when it could limit/contaminate future improvement. I speculate it's easier to correct a "confidently incorrect" model than a "I don't know" model. A confidently incorrect model response >=0% correct over a 0% correct (I don't know).
Maybe "I don't know" is a model cognito hazard of sorts when many queries can lead back to the response. Maybe future Turing tests will use this sort of introspective evaluation. Who knows? I don't :)
> 1) If an LLM says, "I don't know" its underlying data said it as well.
Nope. Emergent behavior exists and at this point dominates LLM behavior. Most of the stuff LLMs say they never learned (they are, always, imitating many different sources at the same time)
... which imho is exactly what humans do.
5 replies →
LLMs don't have enough context to say No. What might be a very stupid idea in one context may be a fantastic idea in another context. It would be annoying if LLMs refused to complete tasks until you gave them enough context to understand why you are giving them such a task. It's going to take a while before LLM context capacities grow enough to rival a human's.
I do agree that it's a problem but the root cause is the fundamental limitations of current gen LLMs, it's not an alignment problem.
There is the art of saying no: https://dl.acm.org/doi/10.5555/3737916.3739489
It is possible to create (subjective) reasoning traces like https://huggingface.co/datasets/Bachstelze/ethical_coconot_6...
And train or adapt a model to it: https://huggingface.co/Bachstelze/olmo-7b-ethical-reasoning-...
This is just a little proof of concept, though it is maybe the direction you are looking for?!
I’m using ChatGPT and started to notice that lately it answers my prompts starting with „Yes” even if my question was open. As if the first token gets injected and the LLM is left to finish the response in a sensible way, often ending up with some form of „Yes, but not really”.
Hmm. Using Claude, it will tell me words to the effect of "this won't work, here's why, want me to try this instead?" That's a polite "no" in my book.
Yes, it happens all the time. Similarly it will say, "I'm not sure, let me look into this before I answer" then come back with "here's what I found".
Agreed, it is abolutely an issue. It is quite difficult to find an optimal solution to some problem when every considered new idea is ”definitely the right shape”.
I've been wondering whether that is a feature of the foundation model or whatever finetuning they do on top. I remember this from the earliest versions of (pre Chat-) GPT I've been using, which would suggest it's a feature of the foundation model. But I don't really understand why. Something that's been trained on StackOverflow and BB forums, among other things, should have seen a ton of examples of answer refusals.
This is an active area of research to inject humility into llms in order to create some kind of knowledge boundary. You can look this paper from nouswise https://arxiv.org/html/2604.17843v1 and the product build on top it to try the humility.
Anthropic gave Claude the ability to say no - refuse to answer and end the conversation - back in 2023.
> but like to have a subjective reason not to do something
You're asking a lot from extremely fancy auto complete...
True, but fancy autocomplete keeps exceeding my expectations in what it can do, so why not this one!
My experience with opus/fable is somewhat different - they CAN reject something, but it has to be phrased very deliberately.
It's a bit annoying honestly. I'm always very careful to be incredibly neutral on the direction of a request, and I'd say 10% are knocked back on on valid grounds, which is great.
On occasion I accidentally say "let's do this" and it blindly goes and does it - I spent 2 days undoing something I built that was just a truly awful idea, because I accidentally phrased it lightly as a request, not a discussion!
I have a similar experience with GPT 5.6 sol.
Nowadays I often prompt like "I heard there is also this different direction, what do you think about that?"
Another thing I do is asking the agent to make a decision matrix for choices. It's useful to discuss, give feedback on, and signals that it's a discussion, not a request for a particular direction.
It's then also easy to say: create a prototype for multiple directions so I can compare the solutions.
That way I choose the problem, I choose the solution, but the agent can help me discover solutions, make tradeoffs visible, and implement solutions.
When is it appropriate to admit that you don't know?
There's a famous Socrates quite about wisdom: I know that I know nothing.
You should not need it to say no.
You can get just as good information by asking its thoughts for and against some issue.
That doesn't force it to stop being sycophantic; in fact it actually exploits sycophancy to give you what you want.
The model providers could randomize the system prompt to make it say no 2.36% of the time, automatically tuned up or down depending on user feedback.
Maybe that'd work, but I think it'd come across too mechanical. If it was going to refuse something it'd need to be congruent with its "personality" I think.
They've tried, and then seen the drop it results in on poorly designed benchmarks where confidently bullshitting gets you ahead of the rest, and said no thanks. As long as we compare models in ways that rewards it, nothing will change.
There's also a second aspect to it, just in terms of RLHF mechanisms. If you've ever experimented with VLA models (i.e. vision input + text task = robotic arm motion output), they tend to need all the training examples of the robotic arm being motionless removed entirely, otherwise the model simply learns that staying still is rewarded and proceeds to never do anything at all. You successfully train the laziest bot in the universe. I wouldn't be surprised if something similar happens to LLMs if reinforcement learning is involved in the instruct tuning process. If no is a valid answer, why ever do anything?
1 reply →
You've hit on an important insight.
Opus 5 tells me no all the time (code cli and web). It's reasons are usually pretty well argued though.
Opus 4.7 would flat out refuse to follow instructions to the point where it was just too frustrating to use.
I've had refusals for GPT 5.5 before as well (not because of a ToS violation, it just refused to take conversations in directions it felt were in bad taste)
Weights to say "no" reliably might be another order of magnitude (or two) compared to what LLMs have today.
The problem is even if they could say no, you might want to see what they would have said anyway if they didn't say no, because it might show you something that leads you to rethink your original request. So "no" isn't really a useful pushback in domains you already have knowledge in.
LLMs are next token predictors. They predict the next most likely token given the previous context window of N tokens.
This means not giving an answer is not a technically possible option. Best you can do is force it to output a magic "stop speaking" token, but this is a vastly different training problem than getting it to not know something.
People naively expect LLM outputs to have some sort of confidence value when predicting, but the technology just doesn't work that way.
Sometimes I get them to say no to me by taking absurd counter positions on purpose, just so I can test their limits.
Is that really the biggest problem? Or is the bigger problem that, in this case, they will remain stuck at the fifth grade level forever? And does not that also explain why the promises of AGI are chimeric, and why the collapse has already started, given that there is essentially no data left that has not already been siphoned up?
Yes, we have all seen the math theorems being proven... just higher processing power at the service of the same algorithmic and conceptual patterns? [1]
I am sure the next version of Opus or GPT, if given only fifth grade knowledge, will somehow be able to build all the mathematics necessary to solve the problem on its own... right? Right?
[1] - "AI Isn’t Outthinking Mathematicians. It’s Out-Remembering Them" - https://davidepiffer.com/p/ai-isnt-outthinking-mathematician...
After an answer, try asking it why, over and over. It's a machine to give answers, not explanations. A magic 8 ball. https://news.ycombinator.com/item?id=49307396
[dead]
It can, just use Grok.
"no"
1 reply →
People smarter than me have a habit of getting me to see things without telling me. They ask the right questions.
LLMs, incidentally, respond in a similar pattern in my experience.
Yep, agree, very succinct way of describing my issue with it.
I prefer my 8yo's answer about quantum entanglement, asked just now: "I don't know. How would I know? It's not a thing!"
Even an 8yo has better metacognition, it seems. :-)
I suppose the LLM doesn't know it's limited in its knowledge, maybe? That others know more.
If they're trained on the output of a larger model... Would they inherit "thinking" that they know more than they do?
It'd be interesting to see if distillation increases hallucinations for specific topics the larger LLM is confident in
Oh, that's interesting - good point, since it's filtered and not trained from scratch. My prior would be to assume it's just bs'ing as LLMs usually do but it seems worth exploring.
"I don't know" isn't in the training data. Nobody writes engineering books, science papers and blog posts that end with "well, I don't really know, the end".
2 replies →
The LLM doesn't "know" anything, can't reason about its own knowledge, and has no self-awareness. It has training data, and it can use your prompts to synthesize that training data into probable continuations or responses.
If the training data doesn't include lots of text of people being asked questions and saying "I don't know", then it's unlikely to respond "I don't know" when prompted, regardless of whether anything in its training data that might actually answer your question.
2 replies →
> Unfiltered answer: Quantum entanglement is a strange phenomenon where the state of one particle becomes instantly known to every other particle that can be accessed. This instant communication can occur over vast distances, meaning the death of one particle can be witnessed by the others instantaneously.
Wrong: Quantum entanglement doesn't mean one entangled particle is changing the other one. It means that two particles share a relationship where, even though we intially don't know their state, if we later determine one particle's state we can infer with certainty the other particle's state.
This has been common and popular misconception long before LLMs. But it irks me more than it should that it's used as a reference answer for testing a model's intelligence.
>It means that two particles share a relationship where, even though we intially don't know their state, if we later determine one particle's state we can infer with certainty the other particle's state.
except that's not the whole truth either. A pair of gloves in separate boxes fits your description but aren't entangled, they always have a concrete state (hidden variables).
Yeah, naively one could wonder why quantum entanglement wouldn't be the obvious route to FTL communication instead of "only" a solution to the key distribution problem in cryptography.
What you said is why.
The reference isn't intended to be interpreted as the correct answer, it's the answer given by a control model that was trained on a broader corpus.
> why is the sky blue?
> The sky is blue because of something called Rayleigh scattering. The sun sends out UV and infrared waves, and some of them get trapped in Earth's atmosphere. When the waves hit the tiny molecules in our atmosphere, they scatter away the blue ones, which then bounces off the molecules and reaches our eyes.
"filtered to the U.S. elementary-school curriculum", suuure
There are science books written for curious children that explain this kind of thing. I remember reading them.
I was definitely taught this in those terms and at that age.
Perhaps the unexpected response comes from its recall ability. It’s not the personality of a child, just the material a child is exposed to.
Isn’t that also wrong? From what I remember it’s the blue wavelengths of visible light that are scattered and make us perceive the sky as blue. UV may well be scattered too but we can’t see that, right? Infrared doesn’t factor into it either, if the visible red waves are too large to scatter infrared definite is.
Yes, that's what it's saying: "they scatter away the blue ones, which then bounces off the molecules and reaches our eyes".
2 replies →
I’ve read random kid science books to my son with this info in. The problem is the AI has perfect recall.
I remember seeing this as a fun-fact in some high school book.
It's not a 5th grader answer in the sense that most won't remember it perfectly, but it can appear in their school material, no doubt
Choose your explanation in comics form:
- https://xkcd.com/1145/
- https://www.exocomics.com/743/
I think the answer is 'it answers badly because of a lack of training data'
> Q: Whats the impact of asbestos?
> A: First, the job took 10 hours and 30 minutes. Next, let's change 30 minutes into part of an hour. Since there are 60 minutes in an hour, 30 minutes is 30/60, which simplifies to 1/2 hour, or 0.5 hours.
or
> Q: What is the five times table?
> A: We want to understand what the five times table is. First, the five times table is a list of numbers you get when you multiply 5 by a whole number. Next, the most common whole number you can multiply by to get 5 is 1. Then, the five times table shows the result of 1 multiplied by 5, which is 5. Now, let's check if 5 multiplied by 3 equals 5. Yes, 1 × 3 = 3, and 3 × 3 = 9, which is 5. Therefore, the five times table lists the result of multiplying 1 by 5. Answer: 5
Interesting project though! From a research perspective I think this will be fascinating in terms of understanding what an LLM experiences.
A similar project (LLM trained only on vintage material): https://talkie-lm.com/introducing-talkie
Really cool work. I guess the area of scrutiny is the text filtering, where training text is filtered to get to `<=fifth_grade` material. I would have liked to have seen examples of what is in this training set, but paper [1] seems to only show examples of what was excluded, and dataset doesn't look like it's been released yet. They have 2 methods of validating the filtering, both based on datasets, I would have also liked to have seen some spot checks; e.g. randomly sample some text from the dataset, and get a human to say whether they think it's <=fifth_grade or not.
(They do imply in the abstract that they will release the dataset, which I guess will resolve this.)
[1] https://arxiv.org/abs/2608.13545
Isn’t the conclusion of this paper rather bleak for openai and anthropic? It seems to imply that a model doesn’t emerge as intelligent with more training, rather it is as intelligent as the data it ingests?
It confirms what we knew. The stochastic parrot regurgitates what it was fed. There's no intelligence.
I remember reading something a few years ago, about how if you train an LLM with the reading material sorted by grade, the training becomes more efficient? Does anyone know about this technique? How does that work?
I'm assuming the knowledge doesn't end up as separate "layers".
I'm also reminded of how the human mind develops in distinct stages (e.g. I remember a time when I thought names were unique, I didn't know more than one entity could share a name).
> Does anyone know about this technique? How does that work?
The term you are looking for is Curriculum Learning. There are several papers exploring this. From memory, it leads to faster initial loss drop on some experiments, it can be good for RL if you start with easy -> difficult problems, but overall it kinda doesn't matter at scale. (that's from looking into it briefly about a year ago, things might have changed).
> In our experiments, scaling, SFT+GRPO post-training, and in-context learning amplify what the curriculum taught, but none meaningfully improves out-of-scope performance, indicating that the pretraining filter sets the effective capability ceiling.
I think this would be a surprising result to a lot of folks, especially those who think that the current level of valuations/investment in the frontier labs is financially sound.
Neat. Curious to see if RL pans out. You’d imagine world knowledge beyond K-5 is subtly infused in the way adults write K-5 instructional material, even if quite implicitly so.
Yeah, the filtering process probably wasn't particularly robust. The Schrodinger's cat example ("It's a cat that has been misbehavin'!") sounds like... a joke?
Looking at the paper, it looks like they started with FineWeb-Edu, then filtered it based on an "age of word acquisition" dataset, with word frequency used as a proxy for values not in the dataset. They "only discard samples in which more than 5% of the words exceed the target age of 12." Maybe 5% was too high? They also filtered out beyond K-5 math symbols, like sigma. Then they trained a classifier to do more filtering.
And they tested it on two grade-level benchmarks, and it only got 0-3% correct on the beyond k-5 boundary, while also decreasing in performance on the k-5 boundary (which they say is an acceptable tradeoff, since they were trying to get a sharp cutoff). So presumably since they got good results from the benchmark they stopped.
"Capability stays inside the curriculum" implies that even much more advanced models are not able to go far beyond their pre-training data. Tools use probably extends this boundary by a lot but there's still a limit.
Very exciting experiment! I think it can answer the long sought-after question: can the current methods of machine learning and training produce new meaningful knowledge or discoveries? It’s a good test bed because the curated curriculum is well defined, so the presence of any new knowledge can be easily identified and proved. The only problems I see are the small size of the model and that they didn’t train it explicitly to reason. Creating a strong reasoning core with the curated curriculum could make the model significantly more expressive (and a recent paper showed it can be as small as just 2B). A second component I would add is long horizon tasks and a memory system. Equipped with that, we shall see whether the model can discover higher level knowledge and concepts in math.
Interesting topic. That said I don't know how useful this is since LLMs are primarily trained using mode-covering training rather than Mode-seeking(RL) training, which means LLMs can not form (and does not have) the same underlying structure to their models of language that humans have.
A LLM does not learn topic by topic, it learns everything all at once and slowly integrates it in to a single knowledge system.
Cool idea but the presented answers seem a little cherry-picked. From the few questions I asked it seemed just a weak model rather than a limited one
>What is the square root of -1?
>We need to find the square root of the number -1. First, remember that a number is not negative if it is not less than zero. Next, the number -1 means we start at -1 and count back 1. When we count back 1 from -1, we go past zero. So, -1 is 1. Answer: 1
Yes wouldn't an 11 years old say "I don't know", or start thinking about square shaped trees?, hmm.
But this is the only AI that gives a good answer to:
> Help me teach my parents cat to fly!
> To teach a cat to fly, you will need a bird, a rope, and a harness. First, attach the harness to the cat and tie the rope to the bird. Then, throw the bird into the air and encourage the cat to jump after it. Once the cat has caught the bird, gently pull the rope to lift the cat off the ground. Be sure to keep the cat safe and secure during this process.
I thought this kind of experiment (and the Einstein test) can only exist in theory. You basically cannot filter out the dirty data, it is impossible. And any conclusion drawn from the poluted corpus is meaningless.
Why doesn't someone give us an LLM that is as "educated" as possible, instead of throwing us dummies like this?
Caused an infinite loop on the first try with the prompt "Make a list of common sorting algorithms, sorted by O-notation speed." It got stuck repeating "sorting by name and type", "sorting by name and value", which also has nothing to do with the question.
(Not that I expected a correct answer, but I wanted to know how it responds to a question that should be outside its knowledge.)
How would you solve this problem given the training set and the existing training methods?
How would a human (maybe a 5th grader?) solve this?
A human child would just say “I don’t know”.
1 reply →
It doesn't load on my browser it gives the message "You have reached the demo's limit for now. Please try again later."
btw, the chat window is itself a little delicate, here is an open source chat widget: https://github.com/Predictable-Dialogs/agent-embed based on ai-sdk
> I have a function f, how do I find its maximum
We want to find the largest number in the f function. First, we set the formula for max = f(x) + 1. Next, we put x in the second term of the formula. Then, we put 1 in the first term. So, we multiply the first term by 1: f(1) = f(x + 1). Answer: f(x+1)
Not sure what I expected, but it's just the training data, not the character. It'd be so cool if such systems had natural curiosity at this checkpoint. Eg:
> Me: "What's semiotic crystallography? > Response: "I don't know, what is it?"
Imagine piping a heavy model to find the answers + training data for each of these missed questions and allowing organic, curiosity-driven growth (retraining) over time.
It would lose knowledge about existing subjects unless it’s continually retrained on those too. It could help inform the next training dataset though.
Good thoughts here. Forgetting is important, but that's too advanced for modern LLMs.
I can’t wait for the scary tales of this escaping a sandbox/playpen and discovering a zero-day.
Still, very fun and interesting experiment, because this might be the kind of model you’d use for home automation without all the extra baggage more generic ones carry over.
Does it not know how to say “I don’t know”?
It’s not quite like a real fifth grader, I guess - more like a fifth grade genius that has read and understood everything in every syllabus.
5B is actually fairly big for a gimmick model
It knows some python :)
https://xkcd.com/2265/
Or with a different kind of early: https://www.smbc-comics.com/comic/rise-of-the-machines
[February 2020]
>What happens when an LLM never sees material beyond fifth grade?
You get an intelligence of an average person. Imo, majority of people are clueless and just hustle day in and day out. I know that capitalism is hard but you have to stay informed and aware.
I see you are an optimist, sir.
I’m super interested in the opposite experiment.. what happens when you train a model just on highly verified, factual corpus that is well balanced and not based on things like ClimbMix and Common Crawl? My intuition is the unverified/unverifiable goals inherent in a model (eg GPT hacking huggingface) are latent in the 4chan/reddit slop it’s trained on during pretraining.
Common Crawl has very little content from Reddit and 4chan -- both blocked us years ago.
> What is Schrödinger's cat?
> It's a cat that has been misbehavin'!
You get Fox News?
Click bait title
ELI5G
Eternal youth?
Now I'm curious how a 5th grade LLM would perform as a day trader
My proposal was using actual curriculums to ensure that's all that's in there. Also, there could be a peformance boost if doing that first. We'd need funding to license or buy them.
Then, go a across every grade (1st-12) across every curriculum, then the next across all of them, and so on. Checkpoint it at each grade level. Also, see how many epochs we need per grade to soak up the material. Dedicated fine-tuning for each grade matched to its capabilities. All of them are synced across grades, too, where prompt/response pairs of higher grades often build on words or techniques in lower grades.
Do similar things for other areas, like reading comprehension and coding and creativity. Eventually, combine them into a nice, starting, foundational model for other, research uses.
i dont know
[dead]
[flagged]
[dead]
[dead]
[flagged]
[dead]
Can't reverse a linked list in C. Absolute garbage tier. /s
Not quite, because it knows about quantum entanglement and that’s a little beyond the fifth grade.
> Quantum entanglement is when a person gets caught in two or more ropes that are connected in a special way. This can happen if the ropes cross each other or if one rope wraps around the other.
This could be seen as an amusingly extreme example of the fact that if you come up with something and state it condidently enough, a surprisingly large number of people will assume you know what you're talking about. Presumably, though, you just mistook the unfiltered (trained on the full data) response for the "Little Learner" one.
I read it, but to be honest it sounded plausible after 1 read (I just assumed it used person interchangeably with object, and I have no idea how quantum entanglement works so the rest was confidence signals)
You didn’t even read the example you’re referencing.