# Neural Networks are Matrix Program Search
>”Neural networks are just matrix program search. We could spend the billions of years performing a search over Turing Machines and we will *never* find one that does something useful, outside of *maybe* adding two numbers.
>
> There is something about the *priors* that we put into these matrices that biases a neural network to learn programs (such as classifying dogs) that are useful to humans. One of the reason could be that quantum mechanics at its core is *just vector algebra*-reality is just made of vectors.” - [Connor Leahy](https://youtu.be/iccd86vOz3w?list=PLwcClAaLqrJy4wrVTB43HDlnzW-d20fUt&t=7330)
---
Date: 20230115
Links to: [Neural Networks MOC](Neural%20Networks%20MOC.md)
Tags:
References:
* [#031 WE GOT ACCESS TO GPT-3! (With Gary Marcus, Walid Saba and Connor Leahy) - YouTube](https://youtu.be/iccd86vOz3w?list=PLwcClAaLqrJy4wrVTB43HDlnzW-d20fUt&t=7330)