Models Are Getting Dumber on Purpose(w4g1.dev) |
Models Are Getting Dumber on Purpose(w4g1.dev) |
So if I'm e.g. coding a SwiftUI app for navigation, I'd take 9B of basic coding and reasoning, add 10B of swift/swiftUI, add 5B of GIS/geography knowledge and another 5B of frontend app design knowledge. My model doesn't need to know a single line of python.
Then when I want to research electronics components, I grab a 15B model of agentic research techniques, and add in 10B of electronics knowledge, etc.
I don't want general purpose models. They try to be everything to everyone. I want to click together a model that is laser-focused on what I am doing, and I want to run it locally
There are "experts" which do divide parts of the model that are found to activate together for specific tasks, so they can be processed in parallel to join the result at the end, but it's nowhere near the granularity of a SwiftUI expert and a python expert. The difference in those things is so trivial from an abstract point of view that it would make no sense. They would be 99% the same.
Distillations also come into this but I'm highly skeptical you could make one guaranteed to only know programming and only in one programming language (especially with as small a sample set as SwiftUI relative to something like C) without its efficacy being hobbled by tunnel vision. Reminiscent of the SpongeBob episode where he empties his mind of everything except fine dining and breathing, then can't remember his name and goes insane. Beyond the basic concepts of general coding and the trivia of syntax, getting anything done requires a large intersection of disparate world knowledge and the ability to apply it to new situations.
Yes, all three together would be even better. But it wouldn’t be if you had 100x more fanfiction, mostly synthetic, generated during RL to teach a model to be better at writing fan fiction. There are real limits to the amount of knowledge you can cram into fixed-size (downstream of hardware availability) weights. For a period scaling with data was basically “free” because we had the Internet and all the books/media that humans had already created; the data was accessible and limited (at least, the parts we think models should know about) enough and top-hardware big enough that we could basically compress the whole thing.
Post-training/RL are making this obsolete because they’re more about skill/capability acquisition rather than knowledge. They can generate much more data (most of it quotidian/useless, ie an agent made a typo in batch 382829) and clearly seem to cause a kind of mode collapse even in the most advanced frontier models.
We don’t need to make LLMs forget about SpongeBob SquarePants so they learn more about bash. But if I have a question about SpongeBob SquarePants, I don’t need to hear about load bearing seams prefaced with honest caveats after a model writes 400 lines of bash to look up SpongeBob’s family.
And there is probably a lot more SpongeBob knowledge we could put into models if we wanted to: interviews with the creative staff, a SpongeEnv/SpongeHarness modeling how the art/story team work together to create entertaining kids tv, a SpongeBench measuring entertainment value, etc. If a SpongeAgent spends 2000 years in Agent University learning how to Spongemaxx we probably don’t need or want to have it spend another 2000 years writing smoke tests
- Understanding of protocols like HTTP.
- HTML, JS, CSS, SVG, and everything "web".
- Understanding of databases, SQL, etc.
- Abstract code architecture patterns.
- Understanding the users' requests in English.
- Responding in English.
- Command line tool usage (agents/harnesses)
- Industry-specific knowledge that can be applied.
- Frameworks, SDKs, applicable libraries.
- Relevant legal requirements.
- Etc...
Instead we ended up with no finetuning. We give audio snippet to 2 AsR models, take 3 best transcriptions and ask the LLm to pick the best based on the context. That produced significantly higher accuracy in how an agent understands the users.
Turns out the world is made of simple, specialist processes, not generalists trying to achieve them. Adaptability may be of great benefit in evolutionary terms or for a walking anthropoid, but the majority of biology, chemistry, and mathematics rely upon specialist process for good reason. See also the old trope about robotics: that's what you call it before it works, otherwise it'd be a dishwasher.
The upshot is: use a generalist to create a simple solution once, and scale that. Don't deploy the generalist at scale, that's a waste of resources and an inefficient solution.
I'd always thought we'd eventually hotload loras or MoE experts.
It would certainly be useful on the robotics/VLA side of things as well; more limited mobile hardware, download and load/unload new skills as needed.
Tbf I also don't really care what facts my models have baked in (for llms at least). I care most that the model understands general logic and then general knowledge of some level is secondary. Reason being is that everything is RAG'd in anyway.
Models spitting out well established facts is cute but I don't really ever want to rely on say "electronics knowledge" that exists in a tenuous and vague form in the model weights.
Humans write books (and datasheets) for a reason. Books are RAG.
If I had to guess, the weights necessary to encode "how to program" are much larger than the final step of "output python."
ie what everyone asking for this fails to immediately realize.
This is roughly what multi-agent systems are built for.
This is possible with models too, but "making one on the fly" is much easier with agent coordination rather than model weights, since they all speak the same language.
There is an IBM Mainframe vs Google Distributed system division here. Like Seymour Cray said - two oxen or 1024 chickens.
Chickens are harder to harness, so a lot of my work is in sled-dog territory for agent harnesses & command structures.
I think I disagree. For some things, maybe that works - but think of a multi-agent system where one agent understands the code, and passes it off to the reasoning agent to figure out what the bug is. This system is going to suck. Because encoding enough info to figure out what the bug is would just be dumping every single line of the code.
So say agent 1 (reasoning) asks agent 2 (swift) to explain what is happening in File.swift. Anything agent 2 passes to agent 1 short of the entire code is a lossy transfer - and then the bug gets missed.
https://linux.die.net/man/1/ls
I think the vast majority of people do want general purpose models. They want to be able to ask it any question, or ask it to perform any task, and for it to do a decent job at it.
I agree that it's really hard (maybe even impossible) to build something that's everything for everyone. But your average (or even above-average) LLM user doesn't want to choose from a catalog to stitch together a model that does just what they need.
I do think for certain domains this is useful and will make sense: the model backing a coding harness doesn't need to know about the politics of 400BCE Rome. But I'm skeptical that many software developers will want to do what you propose, picking knowledge bases that are tailored to their current task or project. And at any rate, for web-based chat interfaces, most users just want to type a query and get an answer.
Part of the problem of this is likely that the deep meanings of words you might use in chat to describe a business problem or task that you wish to see implemented are essentially inseparable from scenarios in which they are used.
Putting aside the bouba/kiki effect and anything like it, complex words only have meanings from usage. That usage is built on grammatical structures that also emerged only from usage.
(This is something I was taught as a sort of fact but I gather it was basically abbreviated Wittgenstein? … who I cannot claim to have studied)
So what you're looking for is a language model where fundamental word meanings are encoded without the weight of knowledge of where they come from. This is plainly difficult, because complex words are used by extension and analogy, and these days, many are neologisms or portmanteaus, even ephemerally — developed and discarded within a single context.
Reasoning about language itself to its full meaning is quite hard.
Like my favourite word of the moment: "obscurantist". You see that and you have a glimmer of what it might convey. But why do you? How much of that comes from explicit grammatical knowledge of suffixes, and how much from simple experience of using words like obscured, informant, attendant, dentist, artist?
So a language model might be able to deduce what "obscurantist" logically means when applied to a tract or to a person. But without lots of parameters covering its use, could it properly grasp that in some circles it would be pejorative to the point of being deeply offensive?
I think the best hope for your pluggable knowledge base idea is model delegation: strong reasoning models that know how to dictate to smaller specialist models and draw conclusions from their responses. I find myself wondering if there's any way that can be done the same way that, say, Gemma 4 12B's integrated vision encoder works — within shared weights, somehow, without them to speak in some intermediate language, like a partitioned brain. But I find it difficult to believe that is pluggable at all.
Another approach would be to have basic coding and reasoning model and then load specification for language and libraries into context, it could work for self-hosted models, but I don't want whole specification of the language to be send to API and waste tokens on that.
THe original MoE paper from Noam Shazeer et al. is worth a read on this bit, though the paper is admittedly pretty dense. But TL;DR is that each expert layer is learning highly abstract, localized structural and syntactic patterns in the data to minimize the loss function, and its doing this token-by-token (which in some cases may have some domain clustering, but that's just incidental).
When you start batching your queries, even if they all seem like theyre in a single domain, if you visualized the activations you'd notice that most if not all of the network is lighting up on the batched forward pass.
And so far even the biggest model doesn't seem to have a working Make No Mistakes module, so maybe that's not needed
Aren't you describing RAG or even MCP servers? Heck, nowadays you get that also with agent skills and specialized tool calling.
Definitely not MCP, as that pulls info into the context. Unless contexts become REALLY big so that I can add 10B in swift knowledge, that's not gonna help me.
Possible RAG? I don't know enough about how that works, but I think that's not quite it either. I don't want to import facts like "the swift standard library contains a reverse array function", i more want to import knowledge - e.g. the parameters used to generate the text to reverse an array in swift.
Tool calling wouldn't do it either. You'd have to encode every single possible bit of useful info into the tool call, and the tool response would have to encode every piece as well (variable names, function scopes, types defined in other files, etc). E.g. how does it find a bug, if you have to pass understanding back and forth between the brain that understands debugging and the brain that understands THIS code?
I think having unused or rarely used weights doesn’t influence the results as poorly as RAG injecting irrelevant facts.
It sounds to me like some sort of “dynamic MoE” where you can add/create or remove experts on the fly.
I think what you’re describing is the closest approximation we reasonably have right now though.
Tell me you don’t know how llm work without telling me you don’t know how llm work. That’s not how they work!
And then ideally, make it pluggable so I can pick what I want from off the shelf components, but if that's not possible - then just train up as many variants as you can so we can all pick the best variant for our current need.
>On SimpleQA, a benchmark of factual recall with no tools allowed, the current leader is Gemini 2.5 Pro at 53%, so the best recall money can buy still misses half the questions.
SimpleQA hasn't been updated in a long time. Gemini 2.5 Pro is a sixteen-month-old model, not "the best recall money can buy".
>The part I find most promising is what this does to hallucination. When a fact lives in weights, a wrong fact is unfindable and unfixable.
This seems confused. LLM hallucinations don't come from the weights containing "wrong facts", they are artifacts that appear at runtime.
>When the fact lives outside the model, a wrong answer has an address. The model cites a document, so you can open the document. If the document is wrong, you edit the document
You can make any modern LLM explain its reasoning and find sources for its claims. None of this has anything to do with facts needing to exist in weights or in harnesses.
The internet is full of wrong information and I cannot magically edit it to make it all correct, so this doesn't help me.
>if a model is factually wrong a claim with a source is checkable and a claim from weights isn't.
Why? If a model's weights claim that Bart Simpson became President in 2020, why does this fact suddenly become uncheckable?
> There's a version of this future where the model card stops listing a knowledge cutoff at all, because what's left in the weights goes stale on a scale of years instead of weeks.
Future?
Even just recently I’ve read of two approaches to this problem:
Cactus have come up with Needle [0][1], which is their tool-calling focused 14 MB model (still an LLM!) – no world knowledge engrained.
And instead of say, tool call structure, VibeThinker [2][3] focuses on reasoning over world knowledge.
Combine these two approaches with a reliable search tool/a safe way of accessing the internet for the model, and you’ve got a probably slightly slower model for factual questions, which on the upside however doesn’t hallucinate.
[0] https://cactuscompute.com/needle
[1] https://news.ycombinator.com/item?id=49246804
To reason properly about the human condition (eg. World War) wouldn't you need to reason on some facts ? And then reason how some "facts" change the human behaviour ? How can you arrive via pure reasoning to predict how a collective of humans act ? We are not reasonable, humans are not logical deterministic machines confined to algebraic rules.
Specifically, creative writing driven by nerds dreaming about a future, without proper grounding in reality, constraints and all that stuff.
Which is kinda ironic given the topic. And also important to do, because we should keep dreaming. We should just also be aware of when we are doing that and mark it as such.
So newer data would be interesting.
(It seems a bit like an AI generated argument that uses old facts - something that happens to me quite often)
There's no reason that an LLM should have a vast number of obscure facts encoded. It can go out to a search engine for such facts. But the LLM has to be clear on what it doesn't know.
(Google's pricing for search from programs starts at $2.50 per 1,000 queries. If an LLM reaches out to Google, it has to pay.)
I wonder what’s tre latest in this field? Did we get a grip on this problem?
That's the right question to ask. For a while, it seemed that hallucinations went down as models got bigger. That may only have been because, with a big enough model, the desired data might be in the model, somewhere, which would keep the model from making up something. That's the brute-force approach to the problem.
This new article indicates that trimming down the model by pulling out seldom used info makes the problem worse again.
If LLMs had reliable "I don't know", and access to search engines, much smaller models might work.
What day, month, and year was Carrie Underwood's album "Cry Pretty" certified Gold by the RIAA?
If your idea of the smartest person in the world is the guy who always wins tuesday night pub trivia, this blog post is for you. It also gets it's foundational factual claim wrong (as seen via epoch.ai). Very on brand.
https://epoch.ai/benchmarks/simple-qa-verified?view=graph&ta...
https://logs.epoch.ai/inspect-viewer/c79c08da/viewer.html?lo...
Current AI is like the film company producing TV series or movies
Your question is like a story outline. You tell the film company that this is the movie you want. The AI film company then searches for existing similar stories. If similar stories do not exist or details are missing, screenwriters use imagination to fill in the gaps (remember hallucination? It's just a makeup.)
So you cannot solve hallucination of AI
For example I prefer Kimi K2.6 1T parameter to Flash V4 0731 230B parameter, even if it is less intelligent.
In terms of the value proposition of AI replacing knowledge workers, all value is in coding agents (coding agents as general agents).
LLMs work is being intelligent not having knowledge of everything is ok. But, they have to be intelligent enough (with some degree of knowledge) that where to find the information (search tools or any other tools for that matter)
That's a sales tactic -- not a logical position.
After that Space colonization will come.
(In retrospect it looks like they were pretty forward thinking!)
Edit: I ran this article through pangram and it is “100% AI generated”. Cool.
Putting readers through this exercise disrespects their time. Even if as a writer you did the work of researching, reasoning, and fact-checking, you shoot yourself in the foot by running it through an LLM because there's no way for the reader to know which thoughts/research are from you. It demolishes the Ethos of the writing; readers feel they must do quality assurance on the reasoning, research, and facts.
The author is factually incorrect here. Moving information out of the model weights and into the input of the model's context window in no way ensures that the model will accurately output content that was input from the context. This is true even when RAG is used to input exactly the correct data.
I'd wager most people have less. In 2022 a 3080 might have 12 GB if you were lucky, 10 if you weren't -- and you paid for the privilege. A current RTX 5080 is only 16GB.
I know is editorialized, but a more accurate title to this content would be either :
Models Are Getting Ignorant on Purpose
or
Models Are Getting Less Knowledgeable on Purpose
> You can make any modern LLM explain its reasoning
You can make any modern LLM create a plausible, self-consistent explanation that looks like reasoning, but it's not "the reasoning it used to arrive at that answer".
We often make a decision based on a gut feeling, and then backfill a logical reason supporting our feeling, without even realizing we're doing it -- rationalization.
Like asking a human "how did you catch that fast ball coming at you?"
>The internet is full of wrong information and I cannot magically edit it to make it all correct, so this doesn't help me.
My favorite RAG experience was asking Bart (or whatever they were calling Gemini back then) an answer to a question I knew.
It gave me the opposite of the truth (as was common with LLMs at the time).
But weirdly, it had cited sources for this "fact."
I checked the sources. Two of them, both AI SEO slop.
In this moment, andai was enlightened...
Providing not just any a baseline, but a correct and useful one, is ever more important the less the model is grounded in world knowledge – misunderstandings probably compound faster if there is no general grasp of (broadly) “life on earth”, or computers, or whatever.
And secondly, I think (consumer-oriented) search becoming worse and worse is a challenge that’s mostly solvable (but far from solved!) for the big labs: (Mostly) trusted or even editorialized/reviewed sources like published work, Wikipedia, etc. is something they could index internally, it doesn’t need to come from a random blog site on the public internet. Furthermore, there’s a whole slew of companies specializing in crawling-for-LLM (i.e., bypassing bot protections) now as well.
And of course the neural network series.
Hearing has volume, direction, pitch, it's spacial processing etc
Sure, 3.8 maybe it's better now, but an accurate comparison would be with a new Gemma4-31B iteration (that doesn't exist).
Assuming you can tweak the training data, regenerate qwen3.6, and get a better coder, then presumably you could have variants - e.g. qwen3.6-swift-27b and qwen3.6-python-27b. Or maybe all coding is too intertwined and you can only get splits like qwen3.6-research-27b and qwen3.6-coding-27b. Which isn't quite my pluggable-models dream, but it's a step closer.
But maybe the difference isn't the training data, it's the architecture, in which case pluggable models is probably not possible.
There is nothing optimal about needing a few billion more parameters to be able to piece together probable answers that can be asserted by querying an oracle.
> I think having unused or rarely used weights doesn’t influence the results as poorly as RAG injecting irrelevant facts.
Those aren't free. The more parameters you add, the higher the computational cost required to train and prompt a mode.
And all for what? To piece together info that you can just query from a data source?
> so LLM performance in swift benefits from pythonic patterns
What I hear you saying is that the best way to make a swift-trained-only LLM smarter is to train it on some python too. And then with an infinite parameter budget, every other programming language or really any other data you train it on makes the model smarter - I accept that premise.
But in a fixed parameter budget, what is better? training on 50% Swift + 50% Python, or 50% Swift + 50% Rust. Because if I am doing Swift programming, I want whichever of the two is better for Swift. If I am doing Rust programming, maybe I want the model trained on 50% Python + 50% Rust. Sure, it would be smarter if you tossed in the swift code too - but we have a budget to stick to.
Now is it possible to make those pluggable? i.e. can you take a model trained on 50% python, and layer on 50% rust OR swift depending on what language you're using? Probably not right now, but maybe one day?
Even basic clients are now harnesses. A lot of chat interfaces are using memory systems, web search and other stuff under the hood.
Not as agentic as openclaw, but not a straight closed conversation either.
It would be trivial to have a pre-flight convo with an llm to guide the user thru module choices. "Build a site" -> "ok, describe the purpose" -> "local council in XYZ location" -> "that implies you won't need localization since XYZ has a monolingual government" -> "english and catalan localization please".
Right now, you prompt and it builds using assumptions, and we prompt to adjust. I think it would be great to be able to pre-load a set of assumptions.
Everyone assumes that carefully crafting a specific AI architecture with bits and pieces bolted together based on their human intuition is necessarily superior to simply using a bigger monolithic AI model. It turns out that the opposite is true, and has been demonstrated over and over again.
The bitter lesson is this: You can simply ask a frontier model to do the things you suggested, in a few terse lines of English. Dump a few lines in AGENTS.md and you are good to go.
Your approach is to "fiddle with inadequate tools" for weeks or months until you can finally attain a pale imitation of what the frontier models can do effortlessly.
It's the classic "But I can customise EMACS endlessly, why would I use an actual IDE?" argument all over.
I get it. You don't feel ownership over someone else's AI. You don't feel involved, you don't feel like you have agency.
It's like LEGO or IKEA furniture: study after study has shown that people enjoy things more if they "put it together themselves", even if fundamentally the thing is worse and/or still essentially nothing more than plastic made in a factory.
I think you're right that current architectures don't compose like that - but I feel like that's a result of the focus on MOAR DATA, and a "race for AGI" - if we set those ideas aside, a more composable architecture seems very possible.