Eliezer YudkowskyVerified account

I'm not sure that's true. Most could, if tbey really really had to. Most people just don't want to.

Spoken like someone who has never personally witnessed somebody fail to learn to code.

If you stuck them in a prison which they had to code their way out of, and had an eternity to do so, do you really think they would never get out?

Literally eternity? Then a monkey with a typewriter could get out. But someone who can't Fizzbuzz might *not* get out because they would literally never think of trying all possible solutions, or devise a valid informal iterator for them if they did.

Actually, the prison analogy works well in poorer countries where everyone learns how to code out of necessity. It's only in free prosperous countries that people aren't willing to force themselves

The category “poor countries that seem to produce a lot of coders” is just a list of poor countries with very large populations. Brains suitable for coding are ~evenly distributed while opportunities are not.

No, they are countries with much closer to even diversity due to the harshness of the economy. When people don't have a choice they prove anyone can learn to do it. Higher than average IQ and passionate interest is all that's needed to learn coding.

"anyone can learn to do it" "Higher than average IQ and passionate interest is all that's needed"

Eliezer - I agree with this, and I think it’s important to say. Related more specific question - which kind of person does MIRI hire? The ones I’ve met seem more focused on writing papers about hypothetical AI capabilities than on creating AI technology.

We are currently looking for Haskell programmers with a good cultural fit.

Why not? (Not saying you are wrong necessarily, but I am just curious how you came to that conclusion and what prevents them from learning it.)

I don’t know why not. I don’t know what it actually feels like from the inside to apply for a coding job and be unable to pass a Fizzbuzz test. But the fact that Fizzbuzz exists as a filter makes it hard for me to imagine someone filtered by it ever being able to learn to code.

I see. I wonder if the test is filtering out people who people who don't have programming degrees/never studied it, or people who actually majored in that (I have a hard time imagining they could get through computer programming classes without being able to pass it).

I’ve interviewed a few CS majors and people with previous coding jobs who failed FizzBuzz or similar. I taught and tutored math for years and am convinced there’s a similar truth: most people cannot learn real math. And it’s not about IQ. I wish it weren’t true.

How can you make it through a CS program without knowing that? I know you tell the truth but I can't wrap my head around it. You are exposed to 100000x the complexity when learning algos/compilers/OS/graphics etc. Even if they cheat, at some point they have to pass the exam, no?

Possibly they relied in school on help from friends and a small set of memorized solutions plus guess-and-checking with print statements? Those two habits will get a person pretty far with minimal understanding. I've seen it done.

If you are intelligent enough to grok natural language, you can and do program on a regular basis. Formal programming languages are strictly lower on the Chomsky hierarchy, so any human NL parser/generator is provably capable of parsing/generating programming languages. QED.