Why AI systems don't learn – On autonomous learning from cognitive science
155 points
16 hours ago
| 16 comments
| arxiv.org
| HN
Animats
8 hours ago
[-]
Not learning from new input may be a feature. Back in 2016 Microsoft launched one that did, and after one day of talking on Twitter it sounded like 4chan.[1] If all input is believed equally, there's a problem.

Today's locked-down pre-trained models at least have some consistency.

[1] https://www.bbc.com/news/technology-35890188

reply
Earw0rm
8 hours ago
[-]
Incredible to accomplish that in a day - it took the rest of the world another decade to make Twitter sound like 4chan, but thanks to Elon we got there in the end.
reply
TeMPOraL
4 hours ago
[-]
This has little to do with the bot, and everything with this being the heyday of Twitter shitstorms; we didn't have any social immunity to people getting offended about random things on-line, and others getting recursively offended, and then "adults" in news publishing treating that seriously and converting random Twitter pileups into stock movements.

In a decade since then, things got marginally better, and such events wouldn't play out so fast and so intensely in 2026.

reply
giancarlostoro
25 minutes ago
[-]
> In a decade since then, things got marginally better, and such events wouldn't play out so fast and so intensely in 2026.

Are you saying the internet would not do it again, or Microsoft would not do the same approach? Because I think the internet would absolutely do it again.

reply
armchairhacker
4 hours ago
[-]
I think models should be “forked”, and learn from subsets of input and themselves. Furthermore, individuals (or at least small groups) should have their own LLMs.

Sameness is bad for an LLM like it’s bad for a culture or species. Susceptible to the same tricks / memetic viruses / physical viruses, slow degradation (model collapse) and no improvement. I think we should experiment with different models, then take output from the best to train new ones, then repeat, like natural selection.

And sameness is mediocre. LLMs are boring, and in most tasks only almost as good as humans. Giving them the ability to learn may enable them to be “creative” and perform more tasks beyond humans.

reply
armoredkitten
1 hour ago
[-]
Exactly. The notion of online learning is not new, but that approach cedes a lot of control to unknown forces. From a theoretical standpoint, this paper is interesting, there are definitely interesting questions to explore about how we could make an AI that learns autonomously. But in most production contexts, it's not desirable.

Imagine deploying a software product that changes over time in unknown ways -- could be good changes, could be bad, who knows? This goes beyond even making changes to a live system, it's letting the system react to the stream of data coming in and make changes to itself.

It's much preferable to lock down a model that is working well, release that, and then continue efforts to develop something better behind the scenes. It lets you treat it more like a software product with defined versions, release dates, etc., rather than some evolving organism.

reply
vasco
6 hours ago
[-]
That one 4chan troll delayed the launch of LLM like stuff by Google for about 6 years. At least that's what I attribute it to.
reply
InfiniteLoup
1 hour ago
[-]
I was always curious about how Tay worked technically, since it was build before the Transformers era.

Was it based on a specific scientific paper or research?

The controversy surrounding it seemed to have polluted any search for a technical breakdown or a discussion, or the insights gained from it.

reply
Kye
1 hour ago
[-]
reply
bsjshshsb
4 hours ago
[-]
Yes I like that /clear starts me at zero again and that feels nice but I am scared that'll go away.

Like when Google wasn't personalized so rank 3 for me is rank 3 for you. I like that predictability.

Obviously ignoring temperature but that is kinda ok with me.

reply
shevy-java
3 hours ago
[-]
> Back in 2016 Microsoft launched one that did, and after one day of talking on Twitter it sounded like 4chan.[1] If all input is believed equally, there's a problem.

Well it shows that most humans degrades into 4chan eventually. AI just learned from that. :)

If aliens ever arrive here, send an AI to greet them. They will think we are totally deranged.

reply
moffkalast
4 hours ago
[-]
Yeah deep learning treats any training data as the absolute god given ground truth and will completely restructure the model to fit the dumbest shit you feed it.

The first LLMs were utter crap because of that, but once you have just one that's good enough it can be used for dataset filtering and everything gets exponentially better once the data is self consistent enough for there to be non-contradictory patterns to learn that don't ruin the gradient.

reply
zhangchen
12 hours ago
[-]
Has anyone tried implementing something like System M's meta-control switching in practice? Curious how you'd handle the reward signal for deciding when to switch between observation and active exploration without it collapsing into one mode.
reply
robot-wrangler
11 hours ago
[-]
> Curious how you'd handle the reward signal for deciding when to switch between observation and active exploration without it collapsing into one mode.

If you like biomimetic approaches to computer science, there's evidence that we want something besides neural networks. Whether we call such secondary systems emotions, hormones, or whatnot doesn't really matter much if the dynamics are useful. It seems at least possible that studying alignment-related topics is going to get us closer than any perspective that's purely focused on learning. Coincidentally quanta is on some related topics today: https://www.quantamagazine.org/once-thought-to-support-neuro...

reply
fallous
10 hours ago
[-]
The question is does this eventually lead us back to genetic programming and can we adequately avoid the problems of over-fitting to specific hardware that tended to crop up in the past?
reply
t-writescode
10 hours ago
[-]
Or possibly “in addition to”, yeah. I think this is where it needs to go. We can’t keep training HUGE neural networks every 3 months and throw out all the work we did and the billions of dollars in gear and training just to use another model a few months.

That loops is unsustainable. Active learning needs to be discovered / created.

reply
exe34
6 hours ago
[-]
if that's the arguement for active learning, wouldn't it also apply in that case? it learns something and 5 minutes later my old prompts are useless.
reply
t-writescode
4 hours ago
[-]
That depends on the goals of the prompts you use with the LLM:

* as a glorified natural language processor (like I have done), you'll probably be fine, maybe

* as someone to communicate with, you'll also probably be fine

* as a *very* basic prompt-follower? Like, natural language processing-level of prompt "find me the important words", etc. Probably fine, or close enough.

* as a robust prompt system with complicated logic each prompt? Yes, it will begin to fail catastrophically, especially if you're wanting to be repeatable.

I'm not sure that the general public is that interested in perfectly repeatable work, though. I think they're looking for consistent and improving work.

reply
naasking
33 minutes ago
[-]
I don't think old prompts would become useless. A few studies have shown that prompt crafting is important because LLMs often misidentify the user's intent. Presumably an AI that is learning continuously will simply get better at inferring intent, therefore any prompts that were effective before will continue to be effective, it will simply grow its ability to infer intent from a larger class of prompts.
reply
aanet
16 hours ago
[-]
by Emmanuel Dupoux, Yann LeCun, Jitendra Malik

"he proposed framework integrates learning from observation (System A) and learning from active behavior (System B) while flexibly switching between these learning modes as a function of internally generated meta-control signals (System M). We discuss how this could be built by taking inspiration on how organisms adapt to real-world, dynamic environments across evolutionary and developmental timescales. "

reply
iFire
13 hours ago
[-]
https://github.com/plastic-labs/honcho has the idea of one sided observations for RAG.
reply
dasil003
15 hours ago
[-]
If this was done well in a way that was productive for corporate work, I suspect the AI would engage in Machievelian maneuvering and deception that would make typical sociopathic CEOs look like Mister Rogers in comparison. And I'm not sure our legal and social structures have the capacity to absorb that without very very bad things happening.
reply
tim333
1 hour ago
[-]
I was kind of worried by them going Machiavellian or evil but it doesn't seem the default state for current ones, I think because they are basically trained on the whole internet which has a lot of be nice type stuff. No doubt some individual humans my try to make them go that way though.

I guess it would depend a bit whos interests the AI would be serving. If serving the shareholders it would probably reward creating value for customers, but if it was serving an individual manager competing with others to be CEO say then the optimum strategy might be to go machiavellian on the rivals.

reply
estearum
1 hour ago
[-]
> I think because they are basically trained on the whole internet which has a lot of be nice type stuff.

Is this not just because their goals are currently to be seen as "nice"?

Surely they can be not-nice if directed to, and then the question is just whether someone can accidentally direct them to do that by e.g. setting up goals that can be more readily achieved by being not-nice. Which... is how many goals in the real world are, which is why the very concept and danger of Machiavellianism exists.

reply
gotwaz
11 hours ago
[-]
Not just CEOs, Legal and social structures will also be run by AI. Chimps with 3 inch brains cant handle the level of complexity global systems are currently producing.
reply
AdieuToLogic
9 hours ago
[-]
> If this was done well in a way that was productive for corporate work, I suspect the AI would engage in Machievelian maneuvering and deception that would make typical sociopathic CEOs look like Mister Rogers in comparison.

Algorithms do not possess ethics nor morality[0] and therefore cannot engage in Machiavellianism[1]. At best, algorithms can simulate same as pioneered by ELIZA[2], from which the ELIZA effect[3] could be argued as being one of the best known forms of anthropomorphism.

0 - https://www.psychologytoday.com/us/basics/ethics-and-moralit...

1 - https://en.wikipedia.org/wiki/Machiavellianism_(psychology)

2 - https://en.wikipedia.org/wiki/ELIZA

3 - https://en.wikipedia.org/wiki/ELIZA_effect

reply
naasking
31 minutes ago
[-]
> Algorithms do not possess ethics nor morality[0] and therefore cannot engage in Machiavellianism[1].

Conjecture. There are plenty of ethical frameworks grounded in pure logic (Kant), or game theory (morality as evolved co-operation). These are both amenable to algorithmic implementations.

reply
qsera
9 hours ago
[-]
https://en.wikipedia.org/wiki/ELIZA_effect

>As Weizenbaum later wrote, "I had not realized ... that extremely short exposures to a relatively simple computer program could induce powerful delusional thinking in quite normal people."...

That pretty much explain the AI Hysteria that we observe today.

reply
ACCount37
5 hours ago
[-]
https://en.wikipedia.org/wiki/AI_effect

>It's part of the history of the field of artificial intelligence that every time somebody figured out how to make a computer do something—play good checkers, solve simple but relatively informal problems—there was a chorus of critics to say, 'that's not thinking'.

That pretty much explains the "it's not real AI" hysteria that we observe today.

And what is "AI effect", really? It's a coping mechanism. A way for silly humans to keep pretending like they are unique and special - the only thing in the whole world that can be truly intelligent. Rejecting an ever-growing pile of evidence pointing otherwise.

reply
qsera
5 hours ago
[-]
>there was a chorus of critics to say, 'that's not thinking'.

And they were always right...and the other guys..always wrong..

See, the questions is not if something is the "real ai". The questions is, what can this thing realistically achieve.

The "AI is here" crowd is always wrong because they assign a much, or should I say a "delusionaly" optimistic answer to that question. I think this happens because they don't care to understand how it works, and just go by its behavior (which is often cherry-pickly optimized and hyped to the limit to rake in maximum investments).

reply
ACCount37
4 hours ago
[-]
Anyone who says "I understand how it works" is completely full of shit.

Modern production grade LLMs are entangled messes of neural connectivity, produced by inhuman optimization pressures more than intelligent design. Understanding the general shape of the transformer architecture does NOT automatically allow one to understand a modern 1T LLM built on the top of it.

We can't predict the capabilities of an AI just by looking at the architecture and the weights - scaling laws only go so far. That's why we use evals. "Just go by behavior" is the industry standard of AI evaluation, and for a good damn reason. Mechanistic interpretability is in the gutters, and every little glimpse of insight we get from it we have to fight for uphill. We don't understand AI. We can only observe it.

"What can this thing realistically achieve?" Beat an average human on a good 90% of all tasks that were once thought to "require intelligence". Including tasks like NLP/NLU, tasks that were once nigh impossible for a machine because "they require context and understanding". Surely it was the other 10% that actually required "real intelligence", surely.

The gaps that remain are: online learning, spatial reasoning and manipulation, long horizon tasks and agentic behavior.

The fact that everything listed has mitigations (i.e. long context + in-context learning + agentic context management = dollar store online learning) or training improvements (multimodal training improves spatial reasoning, RLVR improves agentic behavior), and the performance on every metric rises release to release? That sure doesn't favor "those are fundamental limitations".

Doesn't guarantee that those be solved in LLMs, no, but goes to show that it's a possibility that cannot be dismissed. So far, the evidence looks more like "the limitations of LLMs are not fundamental" than "the current mainstream AI paradigm is fundamentally flawed and will run into a hard capability wall".

reply
qsera
4 hours ago
[-]
Do yourself a favor and watch this video podcast shared by the following comment very carefully..

https://news.ycombinator.com/item?id=47421522

reply
ACCount37
3 hours ago
[-]
Frankly, I don't buy that LeCun has that much of use to say about modern AI. Certainly not enough to justify an hour long podcast.

Don't get me wrong, he has some banger prior work, and the recent SIGReg did go into my toolbox of dirty ML tricks. But JEPA line is rather disappointing overall, and his distaste of LLMs seems to be a product of his personal aesthetic preference on research direction rather than any fundamental limitations of transformers. There's a reason why he got booted out of Meta - and it's his failure to demonstrate results.

That talk of "true understanding" (define true) that he's so fond of seems to be a flimsy cover for "I don't like the LLM direction and that's all everyone wants to do those days". He kind of has to say "LLMs are fundamentally broken", because if they aren't, if better training is all it takes to fix them, then, why the fuck would anyone invest money into his pet non-LLM research projects?

It is an uncharitable read, I admit. But I have very little charity left for anyone who says "LLMs are useless" in year 2026. Come on. Look outside. Get a reality check.

reply
qsera
2 hours ago
[-]
My opinions on the matter does not come from any experts and is coming from my own reason. I didn't see that video before I came across that comment.

>"LLMs are useless" in year 2026

Literally no one is saying this. It is just that those words are put into the mouths of the people that does not share the delusional wishful thinking of the "true believers" of LLM AI.

reply
ACCount37
1 hour ago
[-]
To be honest, I would prefer "I over-index on experts who were top of the line in the past but didn't stay that way" over "my bad takes are entirely my own and I am proud of it". The former has so much more room for improvement.

>Literally no one is saying this.

Did you not just advise me to go watch a podcast full of "LLMs are literally incapable of inventing new things" and "LLMs are literally incapable of solving new problems"?

I did skim the transcript. There are some very bold claims made there - especially when LLMs out there roll novel math and come up with novel optimizations.

No, not reliably. But the bar we hold human intelligence to isn't that high either.

reply
qsera
45 minutes ago
[-]
>my bad takes are entirely my own and I am proud of it"

Sure, but the same could apply to you as well.

>"LLMs are literally incapable of inventing new things" and "LLMs are literally incapable of solving new problems"?

You keep proving that you have trouble resolving closely related ideas. Those two things that you mention does not imply that they are "useless". They are a better search and for software development, they are useful for reviews (at least for a while). But it seems that people like you can only think in binary. It is either LLMs are god like AI, or they are useless.

reply
qsera
4 hours ago
[-]
Mm..You seem to be consider this to be some mystical entity and I think that kind of delusional idea might be a good indication that you are having the ELIZA effect...

>We don't understand AI. We can only observe it.

Lol what? Height of delusion!

> Beat an average human on a good 90% of all tasks that were once thought to "require intelligence".

This is done by mapping those tasks to some representation that an non-intelligent automation can process. That is essentially what part of unsupervised learning does.

reply
reverius42
9 hours ago
[-]
ELIZA couldn't write working code from an English-language prompt though.

I think the "AI Hysteria" comes more from current LLMs being actually good at replacing a lot of activity that coders are used to doing regularly. I wonder what Weizenbaum would think of Claude or ChatGPT.

reply
qsera
8 hours ago
[-]
>ELIZA couldn't write working code from an English-language prompt though.

Yea, that is kind of the point. Even such a system could trick people into delusional thinking.

> actually good at replacing a lot of activity that coders are used to...

I think even that is unrealistic. But that is not what I was thinking. I was thinking when people say that current LLMs will go on improving and reach some kind of real human like intelligence. And ELIZA effect provides a prefect explanation for this.

It is very curious that this effect is the perfect thing for scamming investors who are typically bought into such claims, but under ELIZA effect with this, they will do 10x or 100x investment....

reply
marsten
14 hours ago
[-]
Agents playing the iterated prisoner's dilemma learn to cooperate. It's usually not a dominant strategy to be entirely sociopathic when other players are involved.
reply
ehnto
13 hours ago
[-]
You don't get that many iterations in the real world though, and if one of your first iterations is particularly bad you don't get any more iterations.
reply
naasking
24 minutes ago
[-]
> You don't get that many iterations in the real world though

True, for iterations between the same two players, but humans evolved the ability to communicate and so can share the results of past interactions through a network with other agents, aka a reputation. Thus any interaction with a new person doesn't start from a neutral prior.

reply
cortesoft
11 hours ago
[-]
But AI will train in the artificial world
reply
ehnto
11 hours ago
[-]
They still fail in the real world, where a single failure can be highly consequential. AI coding is lucky it has early failure modes, pretty low consequence. But I don't see how that looks for an autonomous management agent with arbitrary metrics as goals.

Anyone doing AI coding can tell you once an agent gets on the wrong path, it can get very confused and is usually irrecoverable. What does that look like in other contexts? Is restarting the process from scratch even possible in other types of work, or is that unique to only some kinds of work?

reply
Garlef
1 hour ago
[-]
I think restrcicting this discussion to LLMs - as it is often done - misses the point: LLMs + harnesses can actually learn.

That's why I think the term "system" as used in the paper is much better.

reply
troupo
28 minutes ago
[-]
> LLMs + harnesses can actually learn.

No. No, they don't

reply
krinne
5 hours ago
[-]
But doesnt existing AI systems already learn in some way ? Like the training steps are actually the AI learning already. If you have your training material being setup by something like claude code, then it kind of is already autonomous learning.
reply
LovelyButterfly
4 hours ago
[-]
Most, if not all, commercially available AI models are doing offline learning. The cognition is a skill that is only possible on online learning which is the autonomous part the authors refer to, that is, learning by observing, interacting.

In that sense the "autonomous" part you said simply meant that the data source is coming from a different place, but the model itself is not free to explore with a knowledge base to deduce from, but rather infer on what is provided to it.

reply
reverius42
4 hours ago
[-]
> The cognition is a skill that is only possible on online learning which is the autonomous part the authors refer to, that is, learning by observing, interacting.

This is the "Claude Code" part, or even the ChatGPT (web interface/app) part. Large context window full of relevant context. Auto-summarization of memories and inclusion in context. Tool calling. Web searching.

If not LLMs, I think we can say that those systems that use them in an "agentic" way perhaps have cognition?

reply
troupo
26 minutes ago
[-]
No, no they don't. Actual learning survives beyond "sufficient context window".

Start a new chat, and the "agentic" system will be as clueless as before

reply
reverius42
21 minutes ago
[-]
They can write to the filesystem, and future instances can read it (and write more). The agentic system does not remain as clueless as its LLM's first instantiation.
reply
imtringued
1 hour ago
[-]
If you let the AI train on your prompts it will actually learn indirectly. It is still offline learning though.
reply
utopiah
6 hours ago
[-]
I remember a joke from few years ago that was showing an "AI" that was "learning" on its "own" which meant periodically starting from scratch with a new training set curated by a large team of researchers themselves relying on huge teams (far away) of annotators.

TL;DR: depends where you defined the boundaries of your "system".

reply
p_v_doom
6 hours ago
[-]
I think from a proper systemic view that joke is more correct than not. AI is just the frontend of people ...
reply
logicchains
5 hours ago
[-]
There's already a model capable of autonomous learning on the small scale, just nobody's tried to scale it up yet: https://arxiv.org/abs/2202.05780
reply
est
8 hours ago
[-]
"don't learn" might be a good feature from a business point of view

Imagine if AI learns all your source code and apply them to your competitor /facepalm

reply
beernet
16 hours ago
[-]
The paper's critique of the 'data wall' and language-centrism is spot on. We’ve been treating AI training like an assembly line where the machine is passive, and then we wonder why it fails in non-stationary environments. It’s the ultimate 'padded room' architecture: the model is isolated from reality and relies on human-curated data to even function.

The proposed System M (Meta-control) is a nice theoretical fix, but the implementation is where the wheels usually come off. Integrating observation (A) and action (B) sounds great until the agent starts hallucinating its own feedback loops. Unless we can move away from this 'outsourced learning' where humans have to fix every domain mismatch, we're just building increasingly expensive parrots. I’m skeptical if 'bilevel optimization' is enough to bridge that gap or if we’re just adding another layer of complexity to a fundamentally limited transformer architecture.

reply
jdkee
14 hours ago
[-]
LeCun has been talking about his JEPA models for awhile.

https://ai.meta.com/blog/yann-lecun-ai-model-i-jepa/

reply
Xunjin
10 hours ago
[-]
In this podcast episode[0] he does talk about this kind of model and how it "learns about physics" through experience instead of just ingesting theorical material.

It's quite eye opening.

0. https://youtu.be/qvNCVYkHKfg

reply
aurareturn
5 hours ago
[-]
The way I see it, the "world models" he wants to train require a magnitude more compute than what LLM training requires since physical data is likely much more unstructured than internet data.

He raised $1b but that seems way too little to buy enough compute to train.

My bet is that OpenAI or Anthropic or both will eventually train the model that he always wanted because they will use revenue from LLMs to train a world model.

reply
tranchms
11 hours ago
[-]
We are rediscovering Cybernetics
reply
internet_points
3 hours ago
[-]
I've tried figuring out what the big deal about cybernetics was, but I always come away with a feeling of it being a bit wish-washy. Is it a bit like Philosophy in that it birthed individual fields that were inspired by and made applications of the thoughts, models and ideas laid out by its forebears? Or were there actual proofs, discoveries or applications in the field itself?

(I guess one could call projects like https://en.wikipedia.org/wiki/Project_Cybersyn an "application" of its ideas, though cut off before one could see the results.)

reply
walterbell
9 hours ago
[-]
Biological Computer Laboratory (1958-1976), https://web.archive.org/web/20190829234412/http://bcl.ece.il...
reply
QuesnayJr
8 hours ago
[-]
It's striking how cybernetics has gone from dated to timely.
reply
shevy-java
3 hours ago
[-]
The whole AI field is a misnomer. It stole so much from neurobiology.

However had, there will come a time when AI will really learn. My prediction is that it will come with a different hardware; you already see huge strides here with regards to synthetic biology. While this focuses more on biology still, you'll eventually see a bridging effort; cyborg novels paved the way. Once you have real hardware that can learn, you'll also have real intelligence in AI too.

reply
himata4113
4 hours ago
[-]
Eh, honestly? We're not that far away from models training themselves (opus 4.6 and codex 5.3 were both 'instrumental' in training themselves).

They're capable enough to put themselves in a loop and create improvement which often includes processing new learnings from bruteforcing. It's not in real-time, but that probably a good thing if anyone remembers microsofts twitter attempt.

reply
tim333
1 hour ago
[-]
I was thinking in the same way that the human brain's design came about from evolutionary trial and error, we may be close to a situation where we can do something like that for the artificial neural networks and have the computers improve them by fiddling about.
reply
followin_io82
4 hours ago
[-]
good read. thanks for sharing
reply
Frannky
10 hours ago
[-]
Can I run it?
reply
lovebite4u_ai
5 hours ago
[-]
claude is learning very fast
reply