On Sun, Aug 31, 2025 at 9:05 AM Matt Mahoney <[email protected]>
wrote:

> On Fri, Aug 29, 2025 at 10:45 PM Rob Freeman <[email protected]>
> wrote:
> > The contribution of the Hutter Prize to our knowledge has been barren.
> It didn't find the semantic primitives Hutter envisioned for it.
>
> The purpose is not to find semantic primitives. It is to find efficient
> algorithms for language modeling.
>

A casual search pulls up a stated goal of "compressing human knowledge".
Does "compressing human knowledge" differ from finding semantic primitives?

Now "compressing human knowledge" is reframed to be "find efficient
algorithms for language modeling"?

OK.

In practice the biggest change to this compression model in 20 years has
been to allow models to be bigger?

But this is an old argument. Right at the beginning of the Hutter Prize I
used to argue with you that language models would be characterized by
getting bigger, not compression. 10 years later we got LLMs. Not known for
their compactness. But OK, arguments over this go round in circles over
what all the words "mean". There will be some way to argue an LLM is a
compression of something, I'm sure. Yeah, they're a compression of what
they generate, even if they themselves are bigger than what generates
them...

In practice, what we have are LLMs, which get big. No-one has found an
upper bound to the improvement that can be made by making an LLM bigger, so
far, to my knowledge.

The very label LARGE, defines the field.

What they do offer in terms of compactness is something of finite size,
that appears to say new stuff/get bigger.

You can look at the entire history of NNs as the reverse of compression,
really. What's distributed representation? Compression?

So is it a victory for compression to go from symbols to NNs, zero
dimensions to N dimensions is a victory for compression?

And then "deep" nets. More layers. More layers another victory for
compression?

And then the next advance is to go from supervised NNs to unsupervised.
Less structure, or just less imposed structure? Less structure is a victory
for compression? And then from unsupervised to generative...

All the time the field tells itself it is trying to compress stuff, and all
the time what works turns out to be less compressed. We walk eternally
backwards into the future, thinking we're compressing stuff, but in
practice ending up with models that are less compressed.

I guess the question comes down to whether order is necessarily always a
compression. Is the order generated by one of Wolfram's cellular automata
in some way a compression of something?

Anyway, whether seeking compression was a wild goose chase for AI is a
separate argument.

I don't know the OP for this thread was proposing. Sounds like Colin Hales'
electromagnetic field, embodiment, arguments to me. (Embodiment being
another form of a "can't be compressed" argument, BTW.) I'm skeptical of
embodiment too. It takes non-compression to the opposite extreme to my
view. That something can't be compressed, doesn't mean it can only have one
physical realization.

But the argument that AI is necessarily embodied as electromagnetic fields
aside. I felt a need to query the idea that LLMs necessarily equate to rate
based neural models.

I don't think rate based models in neuroscience are settled science at all.
And rate based models certainly don't strike me as a slam dunk argument for
language models in their current form (i.e. LARGE.)

-R

------------------------------------------
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/Ta9b77fda597cc07a-Mcba021e2f34c0476d420ac9b
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Reply via email to