LessWrong Team
I have signed no contracts or agreements whose existence I cannot mention.
Ruby
Karma: 14,329
I think it might well be the case that non-native English speakers gained a benefit from LLMs that native-speakers didn’t, but I don’t think the fact there’s uneven impact means it’s wrong to disallow LLM assistance.
- At worst, we’re back in the pre-LLM situation, I guess facing the general unfairness that some peoplew grew up as native English speakers and others didn’t.
- Practically, LLMs, whether they’re generated the idea or just wording, produce writing that’s often enough a bad experience that I and others struggle to read it at all, we just bounce off, and you will likely get downvoted. By and large, “could write good prose with LLM help” is a very good filter for quality.
- Allowing LLM use for non-English speakers but disallowing it for other usage would be wholly impractical as a policy. Where would the line be? How long would moderators have to spend on essays trying to judge? (but in any case, the result text might be gramatically correct but still painful to read)
- already the moderation burden of vetting the massive uptick in (overwhelmingly low quality AI-assisted essays) is too high and we’re going to have to automate more of it.
It’s sad to me that that with where LLMs are currently at, non-native speakers don’t get to use a tool that helps them communicate more easily, but I don’t think there’s an alternative here that’s at all viable as policy for LessWrong.
(Well, one alternative is moderator’s don’t pre-filter and then (1) the posts we’re currently filtering out would just get downvoted very hard, (2) we’d lose a lot of readers.)
Curated. At first it seems a little odd to recommend lists of content in this day and age when for many or most purposes, you’ll go to your local shoggoth. This compilation is pretty good though. I think there’s something to be said for human works and “cleanliness” of the process that produced them, but also this list contains a number of resources that LLMs just aren’t making yet: some pretty neat visualizations, and e.g. lists of of lists (of lists). An approach I like is to take a reference work and load it into an LLM as context before getting tutoring on a topic. LLMs being able to surface diagrams and images from the text will make this even better.
Here are some of my favorites (though I haven’t look through them all):
Histography by Matan Stauber [interactive timeline]
Landmark Numbers by Miles Kodama [list]
Technology over the long run by Max Roser [chart]
EconGraphs [glossary]
I’m glad this exists and I think is worth taking a look for most people. Kudos!
When I say insurance, I don’t mean it narrowly in the financial sense. I mean it in the “I’ll keep being a part of the relationship even if for some reason you’re less able to deliver on your part of it”, in this case it could be the non-working spouse not leaving when the breadwinner stops winning bread, or whoever sticking around even when you are ill for a prolonged period and much less fun.
Promise to invest, promise to share resources, mutual insurance/commitment seem like other key elements.
Curated. A compliment I’ll give to post is that I’m sympathetic to its length being warranted – it covers and accomplishes a lot, and I could see it being hard to have the same effect with something shorter, akin to how the Meditations on Moloch essay couldn’t obviously be shorter. This too is a meditation, weaving many pieces together such that I think I get the nuanced picture Duncan is painting. At the cost of length, Duncan is very good at covering edge cases and caveats (a software engineer walks into a bar and orders −1 beers, infinity beers, etc). It’d be tedious if everoyne did this but I think it’s valuable that Duncan demonstrates making statements with precision and stemming off misunderstandings and misapplications. (I think “rule out wrong interpretations” being a topic of another of his posts. In in all, I recommend this post.
Now although I haven’t thought about it at length and I’m anxious my judgments are wrong, I do feel a need to purse my lips at the Zack Davis section.
Hopefully fixed now. Also when enabled on the frontpage should work there regardless.
Hopefully fixed now. Also when enabled on the frontpage should work there regardless.
Hopefully fixed now. Also when enabled on the frontpage should work there regardless.
Curated. This is an interesting point to keep in mind – winning control doesn’t mean winning the outcomes you’ve want. I’ve had this thought in terms of getting my way at work (i.e. building the things I want to build or building them in the way I want). Although from the inside it feels like my ideas are correct and/or better, they really have to be for being in control to mean I’ve actually won.
Perhaps in simpler times you win personally if you’re in power (status, money, etc). I think humanity is hitting stakes that yeah, we all win or lose together.
I wonder if a built-in version of aspects of Freedom might be a good idea: i.e. we make it so you can restrict your ability to look at Quicktakes or other sections of the site to an amount you endorse.
Huh, that’s really cut off indeed. This doesn’t happen on any other LessWrong pages?
Oh bummer, which Samsung. Can you share a screenshot here or Intercom?
LessWrong Feed [new, now in beta]
I’ll add that LLMs seem fond of bolding things too and my mind now has “lots of phrases” bolded as a strong heuristic for LLM. Which is unfortunate, because I see the usefulness if it’s well done.
I think having a king at all might be positive sum though, via enabling of coordination.
Curated. The distinction in kinds of power feels reel. I feel like I highlight the “power of having resources” (e.g. money) which is perhaps part of king-power (though John doesn’t list that) or another kind of power.
I wonder if an effect is that living in a large well-developed economy of (i) cheap ready-made goods, and/or (ii) goods that are better due to specialized expertise, equipment, etc., there’s a push against doing things yourself. Even with skills and tools, buying Ikea furniture might still be a better time/money trade-off for most. The barrier to making your better own pain medication than Tylenol is quite high. And so we somewhat correctly get used to exchanging money for things, so we pursue power via having more money in a very abstract way. We also limit ourselves to options being sold to us.
This post then is a reminder that reality can be manipulated directly. Not via commanding or paying others, but via understanding and mastery of reality. Good message. Kudos.
Curated. I don’t think we’re past the point where exposition of how incentives and information flows shape things is valuables. The default lens we/society thinks in terms of is raw power (bigger/stronger/whatever) being the reason for supremacy, and it just feels very instructive to say that just as decisive in how things go is information and motivation. Relatedly, you get things like some places flourishing and some floundering because of different regulatory environments, Goodhart’s. In general, kudos to Arjun for writing that exposes how maybe the actual reason for things isn’t what you’d immediately assume, e.g. perhaps it’s just population size.
My strong guess is this is fine and Eliezer doesn’t mind (and would prefer not to be asked about this kind of thing). Just be clear that it’s LLM generated, who did what, make it clear that Eliezer wrote the underlying essay but didn’t make the music etc.
This is great!
Curated. Simple straightforward explanations of notable concepts is among my favorite genre of posts. Just a really great service when a person, confused about something, goes on a quest to figure it out and then shares the result with others. Given how misleading the title of the theorem is, it’s valuable here to have it clarified. Something that is surprising, is given what this theorem actual says and how limited it is, that it’s the basic of much other work given what it purportedly states, but perhaps people are assuming that the spirit of it is valid and it’s saved by modifications that e.g. John Wentworth provides. It’d be neat to see more of analysis of that. It’d be sad if a lot of work cites this theorem because people believed the claim of the title without checking the proof really supports it. All in all, kudos for making progress on all this.