I think he meant by Tesla using it on the front end that Tesla...

I think he meant by Tesla using it on the front end that Tesla could deploy a trained CNN onto a SNN chip deployed in the car, so that instead of HW3 (the board & chips used in the Tesla car today running the CNN's receiving data from the cameras) they could have a HW? in the future that features an SNN chip supplied by Brainchip.

This would consume less power and make the car more energy efficient whilst still allowing the vehicle autonomously controlled by AI.

DOJO is a massive, industrial AI training infrastructure and it outputs trained NN's

NN's are a bunch of numbers defining the probabilities of stuff (i.e. is that a plastic bag?) & can be deployed onto a chip or board

As Q3 in your document mentioned they move data onto/from chips via NN's

So you can deploy a dumb NN on to an Akida chip & it will learn itself overtime (NOTE: but something still needs to teach it, i.e. when UIUX has to tell it thats a tiger it is seeing. Thats the bit I personally most interested to understand, its the biggest hole in my understanding of Brainchip)

Or you can deploy a smart NN onto an Akida chip & it knows lots of stuff already (trained by DOJO)

I think ......