Internet of Things might have less internet than we thought?

"Think or Thick" cloud architectures.

ML systems have a nice split:
 - development and then deployment.

 There is a shift of inference into the edge and run the models close to the client.


 7% of companies acknowledge getting value/return from their data science or data-lakes projects.

 "You're being watched. Are you okay with that?"

 lots of problems of privacy, security and even compliance w/ law on IoT stuff and "smart gadgets" like roombas, mobile apps, etc..
 software & services of modern tools/assets are still property and owned by the company, not the owner.
 so lots of problems of "accidental" failure due to parent co. stop servicing old products.

 Exposure of the industry failings in regards to change their approach to privacy, security, failure modes, ownership, abandonware.
 the line of argument is that more enclaving and isolation is going to be a necessary direction as we mature our
 tech world.

 The final part is: how moving the models and the code to the edge and making it "self hosted" at the edge.
    massive perf difference, requires TPUs or similar though, aaannd.. quantization of models of course.