Thats sounds like a good manuscript title 🙃

… but my fully distributed model explains 99.9% of the fMRI variance 🥺

In our forthcoming paper, John Hummel and I ask what it would mean for a neural computing architecture such as a brain to implement a symbol system, and the related question of what makes it difficult for them to do so, with an eye toward the differences between humans, animals, and ANNs.

This is great! One question: do you think this analysis extends to brain “decoding” methods?

What does it mean if pure prediction fails? If you get 100% pure prediction you still don’t know how the model predicted unless you run an experiment that manipulates independent variables. A digital clock 100% predicts the time of a cuckoo clock but it works totally differently.

📌

Building larger LLMs to get AGI is like linearly accelerating towards light speed

We then all agree in one respect: There is a lot of hype.  But the problem is with researchers who take the hype seriously?That we should focus on the strengths rather than weaknesses of DNNs? That there is little "confusion that deep neural networks (DNNs) are ‘models of the human visual system’”?🤷‍♂️

Definitely.

A lot of the neuroscience work, particularly in hippocampus, has been focused on content-addressable memory, where data and address are the same. But this might not be the right way to think about memory in the brain. Maybe we have an addressing system that is separate from stored content.