The company (Mythic AI) in the article was featured on a Veritasium video[1] about analog computing.

The interesting part for analog is obviously going to be wear and repeatability, because the digital layer auto-corrects the errors until it becomes damaged and unusable clearly, while analog will slide off somewhat gracefully but you need a way to detect it happening multiple layers later.

AI is somewhat tolerant of errors since, the cumulative error across a lot of biases can approach zero if the errors are random.

[1] - https://www.youtube.com/watch?v=GVsUOuSjvcg#t=15m44s

Mythic shut down recently. Well, technically they still exist on paper, but they're a shell of the former selves. It's a shame, really. Their technology is (was) amazing.

They're back actually - They recently raised a new round of funding and are working to release their next-gen chip.

Source: https://mythic.ai/whats-new/mythic-raises-13-million-to-brin...

IMNSHO, AI will the catalyst for going to nuclear power.

Low power inference chips already exist and digital computing uses exponentially less power every year.

Intel GNA has been in their last three gen CPUs, it is in many places already. I'm not sure if it is low power though.

Can you please provide any sort of justification for your (quite ridiculous) statement. We have quite a long history of signal processing migrating to digital precisely because it is better in every regard, INCLUDING POWER. Maaaaaybe, since AI neuromorphic-like circuits are quite fuzzy, they wouldnt require such strong noise resilience and precision, this could be a potential use case, but there is also a point of reliability and testability, which on these required scales for AI will be completely infeasible for analog circuits.

No. Today you can get a netbook which can be powered by a phone charger.

Because you can push 200W over USB now