2022 BRN Discussion, page-5649

The example you have shown is really good and that is cloud based AI.

There is a space for nuromorphic chip with less power consumption. However when you need a computing power, realtime analysis with huge data sets local processing will not be sufficient whatsoever.

Tesla has some lot's of local processing however they have huge cloud processing where they gather the training data from thousands of hours from individual cars, once they create learnings from that they send it back to those new models to individual cars and the their cars becomes smarter by the day. Essentially your Tesla car is learning from other tesla car being driven. That does not happen in local processing such as what BrainChip is referring.

So chips like Brainchip is used for more trivial stuff i.e. to recognise someone said "hey mercedes" rather then "Hey Julie" compare to driving the car which requires lots of training data from thousands of hours of driving.

For example you can recognise whether there is animal in the photo or test the wine using one shot learning but driving the car does not happen in one shot! it requires the model to be training from thousands of other car driven.