Many companies don't pay ML engineers/researchers for the time spent training an ML model anymore.

I don’t understand this comment. If my model is training for 2 days I expect to be doing other work while I wait for the results. Are/were there engineers/researchers that expect to get paid to do nothing while they wait for paint to dry?

Yes, many of them used the training time for quiet deep thinking without worrying about getting paid.

If they’re doing “deep thinking” about work then it’s literally work. The clock doesn’t stop just because you’re not typing.

Also, many of these positions are salaried so they aren’t actively tracking hours.

Imagine you are an ML consultant with an activity tracker installed. The company is only going to pay you for the short fragments of time you are typing the source code and launching subtasks, but not for the duration of those tasks if your activity tracker shows no activity on your side. I had a client who wanted to negotiate two rates - one for the active part and another one for the training time part (barely covering AWS costs). If you think this is rare then you might be well-insulated from the market.

> I had a client who wanted to negotiate two rates - one for the active part and another one for the training time part (barely covering AWS costs).

This doesn’t make any sense. Training time is billed according as a passed through cost plus, not an hourly rate for the person who started the task.

I don’t know anyone who would have gotten away with billing AWS training time at their own hourly rate as if they were working. That’s ridiculous. Would be hilarious to see someone submit 24-hour days for long training runs though.

I can understand that for outside-working hours tasks. But not for working hours where many companies really try to do the activity tracking-based payments like as if the consultant didn't have time allocated on working on their tasks already.

The goal for the game industry is to not need actresses or actors.

It's been increasingly common in indie game circles to buy libraries of mocap animations and lately facial expressions. Then things like Blender and Unity have been working on amazing solutions to merge various mocaps together.

Several multi-million dollar Unity games have used that process successfully 5 years ago, and can fill in the gaps with tech like unity asset "puppet face". Which works surprisingly well with just webcam. Unity face capture on IOS is also fantastic.