But my guess is that studying applied math and CS would have been better for me per hour than studying science, and the reason I spent that time learning science was largely because I think it's exciting and cool rather than because I endorse it as a direct path to knowing things that are useful for doing alignment research
Strong upvote for this.
Doing things you find fun is extremely efficient. Studying things you don't like is inefficient, no matter how useful these things may turn out to be for alignment or x-risk.
Regarding text, if the problem comes from encoding, does that mean the model does better with individual letters and digits? Eg
"The letter A"
"The letters X Y and Z"
"Number 8"
"A 3D rendering of the number 5"
It feels like a "gotcha" rebuke, but it honestly doesn't seem like it really addresses the article's point. Unless you think GPT-3 would perform better if given more time to work on it?
For that prompt "she went to work at the office" was still the most common completion. But it only happened about of the time. Alternatively, GPT-3 sometimes found the completion "she was found dead". Kudos, GPT-3, you understand the prompt after all! That completion came up about of the time.
Does it really understand, though? If you replace the beginning of the prompt with "She died on Sunday the 7th", does it change the probability that the model outputs "she was found dead"?
From previous posts about this setting, the background assumption is that the child almost certainly won't permanently die if it takes 15 seconds longer to reach them.
Sure, whatever.
Honestly, that answer makes me want to engage with the article even less. If the idea is that you're supposed to know about an entire fanfiction-of-a-fanfiction canon to talk about this thought experiment, then I don't see what it's doing in the Curated feed.
I reject the parable/dilemma for another reason: in the majority of cases, I don't think it's ethical to spend so much money on a suit that you would legitimately hesitate to save a drowning child if it put the suit at risk?
If you're so rich that you can buy tailor-made suits, then sure, go save the child and buy another suit. If you're not... then why are you buying super-expensive tailor-made suits? I see extremely few situations where keeping the ability to play status games slightly better would be worth more than saving a child's life.
(And yes, there's arguments about "near vs far" and how you could spend your money saving children in poor countries instead, but other commenters have already pointed out why one might still value a nearby child more. Also, even under that lens, the framework "don't spend more money than you can afford on status signals" still holds.)
As a counterpoint, here's my experience with NixOS: https://poignardazur.github.io/2021/09/08/nixos-post-mortem/
For example I think many Muslim countries have a lot of success at preventing pornography
Citation needed.
My default assumption for any claims of that sort is "they had a lot of success at concealing the pornography that existed in such a way that officials can pretend it doesn't exist".
I find this comment kind of aggravating.
I'll claim that the very mindset you mention starts with not taking Eliezer at face value when he half-implies he's the only person producing useful alignment research on earth, an that his ability to write an angry rant about hopeless it all is proves that everyone else is a follower drone because they didn't write the rant first.
Like, I think Eliezer deserves a lot of respect, and I'm aware I'm caricaturing him a bit, but... not that much?
I don't even think I disagree with you in substance. The mindset of thinking for yourself is useful, etc. But part of that mindset is to not unironically quote everything Eliezer says about how smart he is.