hustwindmaple1

It's so easy to just create a foreign entity and get access to American Clouds these days ....

"we set up a cost-effective server room by adapting a janitor's closet with ventilation and sound insulation"

That's the spirit! Way to go ~

Would be cool to show some benchmark numbers and the binary size

It's already happening, bro

Very cool work

How does spending money on whatever have anything to do with monopoly?

'Failing the whole way up' is a bit too much.

Echo (and others) was actually successful back in the days, just not in the way Amazon envisioned. Commercially it didn't make sense for Amazon (I guess most ppl don't want to buy stuff over a speaker) but it can still be a great entry point for many other things, given the new AI progress.

There are times when capital is cheap and abundant and there are times capital is scarce and hard to come by

If you have played any large mobile games, then you would not be surprised to see apps downloading massive files during first open.

author is an iOS dev ...

Getting rid of reality labs, then what? Be a sitting duck and getting attacked by Tiktok/Apple left and right? They have basically no moat now.

VR is prob. the only thing that remotely looks promising for META now.

It's quite strange to have a great RS to lead _strategy_

GPT-J is actually JAX

Many ppl don't get it. Commumication is work. And it's hard work. Sometimes way harder than coding.

For engineers, the more senior you get, the more communication you should be doing. In most companies, sitting in a corner and quietly writing code will not get you into director-level, however good you are.

Typical practice for any ads-based companies

Why would s/he? S/he is prob. 10 levels below the actual decision maker and sometimes is not even a fulltime employee.

There are a ton of mobile deployment options that support PyTorch+TF models. It's hard to argue TFLite is the best.

https://github.com/alibaba/MNN

https://github.com/Tencent/ncnn

https://github.com/Tencent/tnn

https://github.com/XiaoMi/mace

Plus, each chip vendor has their own SDK (e.g., Qualcomm has SNPE, MediaTek has NeuroPilot)

It is noticeably faster than the original model (~30-40%) on my machine.

It works just fine.