For context (at least for the first one): https://scottaaronson.blog/?p=...

I genuinely wish the discourse around this issue today were even half as informed and substantive as that blog post.

(spoiler alert: it's not)

Obviously I meant to write 2022, but let's be real: this is all just part of The Long 2020.

@pfau 😂 thanks for the laugh For real though, is there any good, falsifiable definition of consciousness yet? Would have hoped that thousands of years of philosophy would have resulted in at least one. IIT isn't.

@dpkingma No, because consciousness is unfalsifiable.

@pfau We’re in 2022, David.

@syhw Half the reason discussion is so maddeningly bad on this website is that no one can keep track of all the responses, so everyone keeps repeating the same stuff. https://mobile.twitter.com/pfa...

@pfau @tszzl i took a class w daniel dennett and the guy who kept bringing IIT was bashed SO hard

@pseudohex14 @tszzl I'm not a fan of Dennett's approach to consciousness either, but lol.

@pfau @Radegund You could write several columns about the rows that will result from the matrix controversy.

@pfau Press in 2017: Facebook's computers do matrix multiplications and invent creepy new language to communicate with each other, researchers had to stop the matrices from multiplying 🤦‍♂️

@pfau If you know you know

@pfau 2014: Integrated information theory can't be correct - if it were, matrix multiplication would be conscious. 2020: We taught this matrix multiplication to talk

@pfau

@pfau @Flexi23 Since we believe the universe evolves according multiplication by a (large) matrix (quantum hamiltonian), surely this follows trivially? 🤔

@pfau OMFG why did he say this?

@pfau IIT: not just a little bit conscious, arbitrarily much conscious.

@pfau There's a big difference between a *specific* sequence of matrix multiplications being conscious (with some of the matrices being distilled from enormous amounts of training data) and fairly *arbitrary* matmul being conscious, which I think the IIT thing was closer to.

@pfau 🤷‍♂️ We might be expecting too much with the word “conscious”

@pfau Well, looking at the Wikipedia page, IIT went out on a limb. Then fell off.

@pfau The best models in NLP are all about matrix multiplication. Scary

@pfau If matrix multiplication is something realized in an actual causal system, why not? All realized physical systems can represented mathematically somehow. And Aaronson’s blog post was impenetrable; he also pronounces phi “phee” in interviews about IIT. Ugh.

@pfau Damn you just spoiled Vandermonde neural networks (or maybe it's a thing already?)

@pfau Why is matrix multiplication not unconscious? Baysed on an integrated lack of information theory.