[benlivengood]

What I look forward to after research like https://arxiv.org/abs/2603.02491, which demonstrate the necessity of world-modeling capability to achieve satisfactory performance on certain goals, is a refractor the SoTA test suites to demonstrate how much world-modeling is necessary in various task distributions.

There have been a few years now of arguments about the level to which transformers do or do not have a world model (v.s. being purely stochastic parrots like early pre-trained LLMs) and now we have some tools to actually make quantifiable determinations.

[thomastjeffery]

But the stochastic parrot (LLM) is the world model, isn't it? What's the difference?

[simianwords]

Yeah… LLMs clearly already have a world model

[thomastjeffery]

I think it's a good distinction to make between having and being, which seems to be what the whole "stochastic parrots" bit was intended to make all along.

It doesn't make sense to say a model is in possession of its self. That's exactly the sort of poetic anthropomorphization that Bender was criticising here, and a good reason to not refer to an LLM as "an AI".

[simianwords]

Wow this is exactly kind of pedanticism that annoys me with Bender. Glad that this kinda thing is becoming unpopular.

[thomastjeffery]

There is more here then pedantry, but if you aren't interested in it, then go right ahead and live your life.