steenreem

I skimmed the paper but I couldn't figure out what they're doing to make concepts fundamentally different from tokens.

I would think that the purpose of concepts is to capture information at a higher density than tokens, so you can remember a longer conversation or better produce long-form output.

Given that, I would have expected that during the training phase, the concept model is evaluated based on how few concepts it emits until it emits a stop.

I think that considering Amazon, a company with independent business units in completely different business (retail and cloud computing) as an entity with a single personality is about as accurate as thinking everyone from the same country has the same personality.

It's not as if the small group of people at the top determine everything else about the company: they only hire the people directly below them, and their influence decreases the further down you go. I think the larger determinant of a company is the business it is in, since that determines at large what people the company is attractive to and which it tries to hire.

I imagine there's a lot of value in having your employees share a timezone for the benefit of easy cooperation, so it might make sense to have at least timezone-dependent salaries.

There might also be value in having your employees be able to easily physically meet, and then location-dependent salaries make sense.

I'm wondering whether the point you're making can be made even stronger.

I would phrase your argument as simple as: you estimate the world is much more likely to become uninhabitable if we don't use nuclear.

What's the worst case of not using nuclear? It's the same as the best case of using it. Best case, nuclear would enable us to entirely phase out fossil fuel energy production, reducing the world's CO2 production by 25%, reducing green house emissions to the point that large parts of the planet no longer become uninhabitable.

I think much of the software might be the same as at any another company. These companies are so profitable that they're willing to offer more $$$ to get slightly better people to do regular work slightly better than it's done on average.

Shorting a stock and having that information be public seems like a big risk, since it opens you up to people trying to make money by short squeezing you.

Why isn't short trading done secretly?

Isn't YouTube banning certain content exactly what free speech is about? YouTube, a corporate entity with its own goals and beliefs, exercises its right to say what it thinks is the truth. It's similar to most newspapers: they're not entirely objective.

If content publishers want to post Corona related information that YouTube thinks is false, they'll just have to find/create another platform for it.

While reading I wasn't convinced this was going to work due to the collusion problem, until I read the comments on collusion which add an extra condition: "we need votes to be so private that even the person who made the vote can't prove to anyone else what they voted for."

However, this condition is not enough, it doesn't cover the case where someone is being watched while they are voting. The condition should be something like:

"we need votes to be so private that only the voter can know what he voted on"

I don't see many solutions to this, except the '19th century' way of voting at a voting station, or a voting machine that can read minds. Obviously it would be pretty strange inputting your vote through thoughts, especially since this system might never allow you to confirm what you voted on.

However, I guess even ballot voting could suffer from collusion since it's not fundamentally private. You could take a hidden camera into the voting box and record your vote, and afterwards get a payout.

I'd be very interested in understanding mainland Chinese sentiment better. My impression is that there is a lot of love for the current Chinese political system, and I'd like to understand why that is. Can anyone refer me to writing pieces that focus on this?

> After selecting a particular phrase using my cursor control commands, I say “clip [foo]“, and every time I want to enter the same phrase after, I say “paste [foo]“. I therefore only have to dictate a particularly obnoxious variable name once.

A simpler solution might be an editor with an auto-completion feature where completion items can be selected using abbreviations or full names with small errors.