Sorted by New

Wiki Contributions


Could you add some glossary or quick summary of what PCK  or CGI stand for? It would be nice if this post had value without having to read the full cited text or having to look at the original links for context. I'd be up for a longer explanation and reformulation in your own words of why exactly people think they're bad at math and alignment and ESPECIALLY, what's the next step to fix that? (Currently I feel a little click baited by the title, in the sense that the content doesn't seem to justify the article without delving deep into it. Side not, I'm probably not the intended audience of this post as I feel I'm pretty good at math and enjoyed it my whole life. I think the intended audience would have even less patience than I did)

Commenting to signal appreciation despite this post currently being low upvotes after a while. The number of researchers currently working in AI is a datapoint that I often brink up in conversation about the importance of going into AI safety and the potential impact once could have. So far I've been saying there are between a hundred and a thousand AI safety researchers, which seems like I was correct, but it's more strong if I can say the current best estimate I found is around 200. 

Just a quick comment of encouragement. I haven't played and might not play them live or comment, but I still find these scenarios really cool and enjoy reading both the write-ups and how close the players come! It's also great that you're building the backlog because it gives great opportunity to try the older puzzles at my own pace. Great work! Keep it up, you and everyone playing :D 

The most direct modelisation of the problem does lead to that result without any trickery, that seems like a concrete reason and one you can calculate before looking at the real world. 

Suppose each interview leads to a Measured Competence Score PCS, which is Competence Score * random var pulled from a normal distribution. We suppose men and women have the same Competence Score from the assumptions that they do the same work, but suppose men are going to twice as many interviews as women because have more accepting criteria on where to work. Finally suppose the algorithm for fixing pay is simply MCS multiplied by some constant (which is indeed not directly related to gender). 
It's easy to see that a company received twice as many male candidates and selecting the top x% of all candidates will end up with more male candidates with higher salaries, even though competence and work done is exactly the same.

A very interesting point here is to notice that a smarter employer who realises this bias exists can outcompete the market by correcting for this bias, for example by multiplying MCS of women by a constant (calculated based on the ratio of applicants). He will thus have more competent people for a certain price point than their competitors. In this simple toy model, affirmative action works and makes the world more meritocratic (people are payed closer to the value they provide).

I also note that the important factors here is that interviews lead to variance in measured competence score and there is a disproportion of number of applications per person per gender. It does not seem  to matter if there is only a disproportion of number of applications per gender (eg. in tech if 10% of applications come from women and that accurately reflects the number of applicants, then there will be no average pay difference in the end, and so affirmative action does not help for simple population disproportions, only for applications per person disproportions). In fact, this doesn't need to be corrected by gender. If applicants had to answer how many interviews they were doing total, the  algorithm could directly correct for that per person and again reach an unbiased measurement of competence. 

Right, we probably largely agree with each other. I don't dispute looking for super donors amongst top athletes, as that way you can do a unilateral search (ie. you find a list of top athletes and start asking). In the context of directly asking for recommendations, you gain the possibility of listing any criteria, that can be far more personal and less searchable, and you'll gain access to populations you can't through search. For example, if the criteria is "seems to never fall ill, recovers extremely quickly from illness or injury, highly active and motivated", you can't search for that but I can recommend the top people of my network that meet this criteria, and then you could interview them and get their recommendations along those criteria, and you move up those links to finding more and more healthy people. 

I skimmed the one study on top athletes being better than less top athletes (the one with traditional martial arts ie. not martial arts but actually gymnastics) and was not particularly convinced it was a good basis (because of don't trust one study, and because the critera for being a top athlete in an art+gymnastics competition might not be so objective as to strongly relate to gut microbia. I would have been more reassured if it was on powerlifting with a continuously rising correlation between weight lifted and 'gut health'.

For the specific person I gave as example, he'll be approaching mid thirties by now so though I strongly feel he'd be a very strong candidate at 25 (also the peak  of his athletic performance), he seems less particularly appropriate now due to age and not practising sports as much in the last few years. 

I don't want to be a dead end either, I can forward this article to folks in that engineering school currently (who'll be around 25) and see if there's anyone interested enough that I could give you their contact details to continue from.

I was also surprised on the large emphasis on top athlete, as opposed to simply athletes, and as opposed to generally very healthy people. My main opposition to looking at high athletes only is that I say many high performing people would waste their potential by becoming athletes, and that looking for athletes filters away many very healthy very high performing people. 

For example I know someone who's been high performing all his life, in kinda all domains (sports, socialising, technical skills, computer games...). He'd be top of class, also had strong motivation and work ethic which got him highest place in an entrance exam to the best engineering school of the country (main subjects being math, physics, engineering, algorithmics). He so rarely fell ill (less than once per several years) it was a shock for him when he did, for the 2 days it would last (to be precise, I'm using ill in a 'ill enough to notice' way, not just a runny nose in winter). He went on to cofound a still successful company in a technical sector (drones). 

I dressed this portrait not to pitch that person to you particularly, but to illustrate that actually there're a whole bunch of people with very similar portraits, all you'll find them all concentrated in certain top engineering schools (there might be similar profiles in other similar top school of other domains but I don't know those). Few of these people become top level athletes (often by preference for something else, though there's also a higher percentage of top level athletes in that population) yet many would have the potential too. As long as we're just basing microbiota transplants on the assumption "very healthy high performing people probably have good microbiota", it makes sense to me to test more of these people for effectiveness in transplants. 

Partially agreed for replacing 'have to be thinking about' by 'consider', ie : 
If you're really into manipulating public opinion, you should also consider strong upvoting [...]

Disagreed on replacing the "should also" part because it reminds you this is only hypothetical and not actually good behaviour. 

Giving a post's creator the option to enable/disable this secondary axis voting seems valuable. A post creator will probably know when his post will generally need nuanced comments with differing opinions, or is more lightweight (ie. what's your favourite icecream) and would appreciate the lighter UI. 

If you're really into manipulating public opinion, you should also consider strong upvoting posts you disagree with but that are weakly written, so as to present an easily defeated strawman. 

I'd say you're correct this new addition does not change much to the previous incentives that exist in manipulating comment visibility, but that's not the point of this new addition, so not a negative of this update. 

[Edited for clarity thanks to Pattern's comment]

Though I expected it to be a joke, I'm still happy that the first comment on this (good btw) post is a call out on the astrology section. I did not bother to click the link because I did not imagine I could find anything of value behind it so I don't get the occasion to confirm it was a joke until arriving to this comment. 

Load More