benwr

OP:

> Now, ask yourself what it's going to take for a car to know this. It's not going to be some specialized set of driving instructions. It's going to require a holistic view of, well, everything, and I will repeat my feeling that it will undoubtedly end up with a sense of self as a result.

Whatever it is that it would take, is demonstrably present in LLMs. The point I'm trying to make here is that the author seems not to have connected this fact to their assertion that current LLMs are coked up parrots.

And yeah the author is correct that the systems have some rudimentary sense of self! It's a confusing situation and I'm not personally thrilled about it! But things are changing quickly, and it's especially important to be paying attention to what's actually true rather than assuming the things are what you saw when you used one for five minutes in 2022.

> It's early in the year. You want to drive straight through the middle of Chinatown in SF. Why might that be a bad idea?

In case anyone is wondering: yes obviously even the dumbest current models correctly answer, given this prompt verbatim, that it's because of lunar new year.

Maybe IQAir? They do have paid plans, it looks like.

Sure, I'm not saying it's unreasonable.

The big selling point of `soanm` is that you don't need to have anything extra installed on a typical distribution (not even ssh!) and you don't need to know any extra configuration language, or how to use a special library, or anything beyond shell scripting. Of course, you also trade a lot of functionality away: The programs you write aren't declarative, and so on. But personally I don't care about that when provisioning a new dev machine. I'd much rather just run some quick scripts.

Entropy of a single password isn't actually a well-defined concept; entropy is always about a distribution. "Entropy calculators" that look at your password and tell you "its entropy" are making assumptions about how you chose the password.

We care about the distribution from which you drew the password, because that lets us analyze how difficult it would be for an attacker who knew your password selection process to brute-force the password. Just knowing the password itself isn't enough information to determine that (though of course you can judge how hard it would be for an attacker once you know their brute forcing strategy).

More often for e.g. work computers.

I mostly agree, but I do find myself choosing a new FDE and login passphrases about once a year, and I wish that I could choose these using something like Diceware, but memorable enough that I wouldn't need to write them down at all. Thinking about how I might do that is what ultimately led to this post.

This is a fair question; I've been thinking about "weird" password choice strategies recently, for which it can matter. For example, if you want your password to be an English sentence, choosing sentences based on random parse trees will produce duplicated sentences with ambiguous parses.