[elcritch]

Yes, but Parrots don't mimic the stages of learning speech development like children do. They just memorize a phrase.

These SOTA LLMs aren't trying to mimic existing children's drawings, but interestingly they're following somewhat similar progression that human children do as they develop.

[diegof79]

They don’t develop. We should stop anthropomorphizing LLMs.

AI labs are improving the ML techniques used to build better models; it’s a big difference.

[perching_aix]

That's very clearly what they're referring to also, or at least I certainly doubt this would be lost on them, or most anyone here, at this point.

[diegof79]

I imagine that it should be clear to everyone here now, until I read threads like this comparing child development to the advances in LLM building.

It’s like saying that a new M-Apple chip developed like a child. There is a clear tendency to give human like attributes to LLMs, probably caused by the terms used in ML: training, learning, etc.

[perching_aix]

I'm not so sure about this view.

For me, it was the same genre of comment as e.g. comments that call out how vision models tend to be structured surprisingly similarly to how the visual cortex is structured, even without specifically attempting to make that happen, and within the boundaries of our understanding of either things of course.

So not so much anthropomorphization, and more a recognition of the unintentional, and surprising, biomimicry.

Though to be honest, I'll be damned if any of the pictures in the post look anything like any child drawing I've ever seen, Grok's or any of the other models.

[elcritch]

No, it's more like you're being overly literal. As a field the technology is developing. Just like people might say "in the 60's integrated circuits developed quickly".

The parent wasn't referring to individual LLMs developing, but that overall new SOTA LLMs models are increasing in capabilities in ways similar to how human children do.

[ETH_start]

Isn't LLM training, the development?

[ipaddr]

LLMs come complete they don't train on the fly based on interaction with the world. LLMs are wikipedia if they stopped allowing edits and all of the previous edits are training.

[saidnooneever]

there is infact RL for LLMs and its unlikely the frontier labs do not use it in some ways

[tadfisher]

The LLMs themselves are not doing the developing or learning.

[ETH_start]

Can you explain what you mean by this? Aren't the LLMs developing as they're trained?

[ben_w]

Both statements are true without contradiction.

A specific model develops as it passes through training.

AI labs change the architecture between models to allow them to surpass the previous models' best scores.

It's also going to keep being a hard sell to say "stop anthropomorphizing LLMs" when the models anthropomorphise themselves.

But besides that: Yeah, sure, they're not human, they're a cargo-cult mimicry of by and of minds that popped out of evolution doing gradient decent on intergenerational survivability. So what? Still interesting when the result of such cargo-culting incidentally echoes what our natural-selection-not-engineered moist electrochemistry happens to do.