"Lack" isn't the right word. "Lacking" is more like it.

If there was a deep fundamental inability, we wouldn't see things like newer generations of LLMs consistently improving on ARC-AGI series (heavy spatial reasoning loading) and SimpleBench (a lot of commonsense + spatial reasoning components).

In a way, it's a surprise that LLMs, notoriously lacking any sort of embodied experience, can even get this close to human baselines on tasks like this.

My takeaway is that text is a far richer modality than anyone has expected - and that high end LLMs are often sharp and flexible enough to recognize their weak points and substitute their strengths. I.e. all the LLMs implementing A* to optimally solve pathfinding in ARC-AGI-3 tasks, often unprompted.

There might still be unrealized gains there from true depth-unbounded recurrence, or maybe from finding better ways to integrate modalities in training. But clearly, a "fundamental limit" it ain't.

▲

fidotron 3 hours ago | parent [-]

> "Lack" isn't the right word. "Lacking" is more like it.

Yeah, that's fair.

> My takeaway is that text is a far richer modality than anyone has expected - and that high end LLMs are often sharp and flexible enough to recognize their weak points and substitute their strengths. I.e. all the LLMs implementing A* to optimally solve pathfinding in ARC-AGI-3 tasks, often unprompted.

I agree and disagree with this. I think we've learned a lot of humans are more text based than we thought, but conversely I'm not persuaded what non-textual task reasoning LLMs are doing is necessarily text based, just that models have grown large enough for other reasoning modes to conceivably be hiding in the parameter space.

As I mentioned elsewhere, like many others I find LLMs work entirely by example, and reaching for A* when pathfinding is the single obvious thing to do. In cases where the magic key word is not mentioned and the problem cannot be identified as "pathfinding" (or some other trigger with a highly specific widely documented solution) they will struggle, yet the moment the trigger is hit they get there very fast. This is why prompting remains such an art form.

Fable is the first one I've encountered that is capable of serious open ended 3D programming in ways that suggest it has some grasp of the spatial aspects of the problem (not merely symbolic manipulation of the vectors etc.), but it still misses optimization opportunities a human will find glaringly obvious based on spatially predictable bounds etc.

	▲	ACCount37 2 hours ago \| parent [-]
		Grown? LLMs were always "large enough for other reasoning modes to conceivably be hiding in the parameter space". Basic LLMs don't reason in text, and never did. They use it as an interface - for input, output and some of the intermediate products. Heavy use of those "pseudo-recurrence" intermediates in "reasoning models" is a relatively late post-training adaptation. But the process that happens between those endpoints is not at all text-based. What happens in the hidden dimension is part "output logit domain", tied to probability distributions over possible output tokens, and part "incomprehensible concept-space madness". The latter being where things like latent world models live. LLMs develop partial world models, right in pre-training, despite not being explicitly forced to - because it brings them closer to heaven of accurate next token prediction. And yes, larger models like Fable seem to be better at spatial reasoning. Maybe because their large size increases the sample efficiency and improves generalization, allowing them to absorb the sparse signal of "spatial reasoning" in the training text better. Maybe because this extra size means more layers, allowing for deeper latent space reasoning in lieu of true recurrence. Maybe because the default "next token prediction" reward underrates rare spatial reasoning challenges, and the model only starts to "get good" at them once the other sources of loss reduction are heavily depleted. Maybe because no true recurrence is suboptimal for spatial reasoning architecturally. But it is what it is. Spatial reasoning gains in LLMs are extractable, but extracting them is nontrivial.