Models Are Getting Dumber on Purpose

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.

ie what everyone asking for this fails to immediately realize.

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

https://linux.die.net/man/1/find

https://linux.die.net/man/1/bash

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

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.

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.

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?"

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.

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.

After that Space colonization will come.