Comment by toisanji

> From reading that, I'm not quite sure if they have anything figured out. I actually agree, but her notes are mostly fluff with no real info in there and I do wonder if they have anything figured out besides "collect spatial data" like imagenet.

Right. I was thinking about this back in the 1990s. That resulted in a years-long detour through collision detection, physically based animation, solving stiff systems of nonlinear equations, and a way to do legged running over rough terrain. But nothing like "AI". More of a precursor to the analytical solutions of the early Boston Dynamics era.

Work today seems to throw vast amounts of compute at the problem and hope a learning system will come up with a useful internal representation of the spatial world. It's the "bitter lesson" approach. Maybe it will work. Robotic legged locomotion is pretty good now. Manipulation in unstructured situations still sucks. It's amazing how bad it is. There are videos of unstructured robot manipulation from McCarthy's lab at Stanford in the 1960s. They're not that much worse than videos today.

I used to make the comment, pre-LLM, that we needed to get to mouse/squirrel level intelligence rather than trying to get to human level abstract AI. But we got abstract AI first. That surprised me.

There's some progress in video generation which takes a short clip and extrapolates what happens next. That's a promising line of development. The key to "common sense" is being able to predict what happens next well enough to avoid big mistakes in the short term, a few seconds. How's that coming along? And what's the internal world model, assuming we even know?

> Here is a summary paper I wrote discussing how the entorhinal cortex, grid cells, and coordinate transformation may be the key: https://arxiv.org/abs/2210.12068 All animals are able to transform coordinates in real time to navigate their world and humans have the most coordinate representations of any known living animal. I believe human level intelligence is knowing when and how to transform these coordinate systems to extract useful information.

Yes, you and the Mosers who won the Nobel Prize all believe that grid cells are the key to animals understanding their position in the world.

https://www.nobelprize.org/prizes/medicine/2014/press-releas...

I kept reading, waiting for a definition of spatial intelligence, but gave up after a few paragraphs. After years of reading VC-funded startup fluff, writing that contain these words tend to put me off now: transform, revolutionize, next frontier, North Star.

Thanks for your article. The references section was interesting.

I'll add to the discussion a 2018 Nature letter: "Vector-based navigation using grid-like representations in artificial agents" https://www.nature.com/articles/s41586-018-0102-6

and a 2024 Nature article "Modeling hippocampal spatial cells in rodents navigating in 3D environments" https://www.nature.com/articles/s41598-024-66755-x

And a simulation in Github from 2018 https://github.com/google-deepmind/grid-cells

People have been looking at spacial awareness in neurology for quite a while. (In terms of the timeframe of recent developments in LLMs).

What I personally find amusing is this part:

>3. Interactive: World models can output the next states based on input actions

>Finally, if actions and/or goals are part of the prompt to a world model, its outputs must include the next state of the world, represented either implicitly or explicitly. When given only an action with or without a goal state as the input, the world model should produce an output consistent with the world’s previous state, the intended goal state if any, and its semantic meanings, physical laws, and dynamical behaviors. As spatially intelligent world models become more powerful and robust in their reasoning and generation capabilities, it is conceivable that in the case of a given goal, the world models themselves would be able to predict not only the next state of the world, but also the next actions based on the new state.

That's literally just an RNN (not a transformer). An RNN takes a previous state and an input and produces a new state. If you add a controller on top, it is called model predictive control. The most extreme form I have seen is temporal difference model predictive control (TD-MPC). [0]

[0] https://arxiv.org/abs/2203.04955

The question, as always, is: can we get any useful insights from all of that?

Trying to copy biological systems 1:1 rarely works, and copying biological systems doesn't seem to be required either. CNNs are somewhat brain-inspired, but only somewhat, and LLMs have very little architectural similarity to human brain - other than being an artificial neural network.

This functional similarity of LLMs to the human brain doesn't come from reverse engineered details of how the human brain works - it comes from the training process.

This is super cool and I want to read up more on this as I think you are right insofar as it is the basis for reasoning. However it does seem more complex than just that. So how do we go from coordinate system transformations to abstract reasoning with symbolic representations?

> if they have anything figured out besides "collect spatial data" like imagenet

I mean she launched her whole career with imagenet so you can hardly blame her for thinking that way. But on the other hand, there's something bitter lesson-pilled about letting a model "figure out" spatial relationships just by looking at tons of data. And tbh the recent progress [1] of worldlabs.ai (Dr Fei Fei Li's startup) looks quite promising for a model that understands stuff including reflections and stuff.

[1] https://www.worldlabs.ai/blog/rtfm

No you just don't understand, don't you see! the ancient greeks foresaw this centuries ago, we are just on the cusp of a world changing moment can't you feel the buzzwords flow through you! First it's creating 7 second meme videos w/ too many arms, then it's right to curing cancer and solving physics! Let the power of buzzwords calm your fears of a bubble.

To decipher if there is anything like spatial intelligence, which is an oxymoronic term at most and redundant at least, one has to decipher the base units of the processes prior to their materialization in the allocortex. And to assign a careful concatenated/parametric categorization of what is unitized, where the processes focus into thresholds etc. This frontier propaganda and the few arvix/nature papers here are too synthetic to lead anywhere of merit.