is it possible to distill a large model into a (even) smaller MoE model, like OLMoE?

You are underestimating...

I would agree. I'm the defendant in a (frivolous) trademark infringement case and my firm (which does work for the EFF) averages about 2 hours per (substantial) page, at a rate of $500/hour. I do wonder in the back of my mind if I'm being taken to the cleaners...

I have a limited experience, only having been involved in one trademark case professionally and am not a lawyer, but this was a case that very quickly settled and fees were still close to $30,000.

this is a great idea and should have happened long ago..

https://news.ycombinator.com/item?id=19713604#19714732

related practical application

https://www.beckman.com/liquid-handlers/echo-acoustic-techno...

Windows 2000 was better

you're all going to upset the actual professional engineers

Have you been to McDonald's lately?

Yes, only robot is a LCD touch screen that replaced the cashier. But what about the robot burger flipping machines. I have yet to see or hear any. Something like these: https://www.latimes.com/business/technology/story/2020-02-27...

Regarding food and cooking: considering that the ice-cream machine at McDonald's breaks down so easily, and is so difficult/costly to repair, one shudders to think of the operational cost and hazard of a robot cook in the kitchen.

I have my doubts about how often those machines really break down.

If you ever want a fun adventure try to get a milkshake in Phoenix at 3AM in the middle of summer. All the machines are conveniently “broken”.

I would love a way to flexibly view and move around lots of PDFs at different scales, to simulate piles on a desk or the floor. For some purposes, something is lost by constraining organization to a regular grid or table.

What ever happened with Nantero?