For people who do test prep seriously (I used to be a full time tutor), this has been known for decades. One of the standard things I used to tell every student was if you have no idea what the answer is, guess B, because B is statistically most likely to be the correct answer. When I was in 10th grade (this was 2002), I didn't have anything to gain by doing well on the math state standardized test, so I tested the theory that B is most likely to be correct. 38% of the answers on that test were in fact B.

> This is pretty weird. As far as we know, humans don’t tend to prefer choices labeled B, so we’re not sure where this could have come from in the training data. As humans, it initially didn’t even occur to us to look for it!

Remember, LLMs aren't modeling how a human reading text would process the text. LLMs are trying to model the patterns in the texts that are in the training data itself. In this case, that means they are doing something closer to imitating test writers than test takers. And it is well known that humans, including those who write tests, are bad at being random. 

Great post! 

> a client can show me where they buried their dozen murder victims and I wouldn’t be allowed to tell a soul, even if an innocent person is sitting in prison for their crimes.

For any Alaskan serial murderers reading this, do note that this does not apply to you. Your Alaskan attorney can breach attorney-client privilege to prevent an innocent person from going to jail. See Rule 1.6(b)(1)(C).

I am talking about the community norm of not using lawsuits to settle arguments (and, more generally, disputes that are… just about words, let’s say). It’s not exclusively a property of small communities—that’s my point.

Then you misunderstood what I was trying to point at when I brought up the distinction about the scale of a community. I was trying to point at the fact that when lawsuits occur, there is likely already too much negative feeling between the parties for them to enjoy direct interactions even in a small group context, and if there isn't already, the lawsuit causes it. I was pointing to a fact about human psychology, which on a pragmatic level we need to arrange our social structures to deal with. I was not pointing to a norm about using libel lawsuits.
 

At this point, you've failed to engage with my point that having a norm against lawsuits is harmful, even though lawsuits themselves are also harmful. I'm guessing your case for having such a norm is that lawsuits are harmful, which is not something I dispute. Is there any more to say?
 

But the fact that we’ve not deviated from the behavior the norm would mandate, is evidence of the norm’s effective existence.

Negligible evidence, especially in comparison to the lack of any past discussion of such a norm. Your argument here is so bad, and your choice of language so ambiguous, that I have to question whether you are even arguing in good faith.

Can you articulate what exactly the property of small communities is that we are talking about, and what its benefits are? I still am not forming a coherent picture of what the heck you are talking about because, again, the thing I was trying to point to in making this distinction I think is inherently a property of small groups.

Are you seriously now claiming that all of society has a norm against lawsuits? I think that is just obviously wrong, particularly for the US. And the misappropriation of the more traditional "arguments get arguments, not bullets" is just astoundingly oblivious. Lawsuits are a kind of argument! They are an example of the thing we are supposed to do instead of bullets!

No, I cannot empirically observe that the rationalist community has operated by such a norm. I can empirically observe that I know of no instance where one rationalist has actually filed a libel suit against another, but this is much more likely to be due to either (1) my ignorance of such a suit, or (2) the low rate of actually filing libel lawsuits in society at large combined with the small size of the rationalist community. I know of no instance of a rationalist going to space either, but I'm pretty sure we don't have a norm against it. I'd never heard anyone speak of such a norm until the NL drama. That is significant evidence that there is no such norm.

May I ask which city you live in?

I agree that these different sorts of communities exist along a continuum. What startles me is that you seem to think that the intimacy of the something of the smaller community can and should be scaled to the larger sort of community. To my mind, it is inherently a property of the small size. Trying to scale it sets of loud alarm bells. I'm not sure to what extent I endorse this, but possibly one way of summarizing the problems of overly controlling organizations like cults is that they try to take the intimacy or something of a small community and scale it.

I also strongly disagree with your presumption that we are talking about "Going from having [the understanding that we do not use libel suits within the community], to not having it". I have never understood the rationalist community to have such a norm. From where I am sitting, Habryka is trying to create such a norm out of nothing, and I am not ok with that.

As I believe I have said already, I agree that libel suits, and law suits generally, can be damaging, and I certainly do not encourage anyone to use them. I'm just pointing out that having a norm against using lawsuits can be even more damaging.

A real court would apply complex rules of evidence, which sometimes involve balancing but often are more categorical. But yes, it's a different notion of public interest than whatever one rando thinks is public interest.

I agree that there is a significant difference between cases where the accused knows the identity of the accuser and cases where they do not, and we should split our analysis.

In cases where the accused does not know the identity of the accuser, I think the accusations would necessarily be so vague that I wouldn't update much on them, and I would hope other rationalists and EAs wouldn't either, but clearly there is a significant contingent of people in these communities who do not share my epistemic scruples. Given that, I don't know, seems a mess. But your rule that only the accused should share the identity of the accuser seems too absolute - surely accusers are sometimes in the wrong, and sometimes malicious, and in that case having their identities publicly known seems good. Yes that will result in some amount of social punishment, and if the accusations were false and malicious, then I think that is good.

The case where the accused does know the identity of the accuser is where my above logic about the accused appearing retaliatory would suggest it is better for a third party to name the accuser.

I think you are using an inapplicable definition of "community". Your example of a D&D group calls to mind a "community" in the sense of "a group of single digit number of people who are in the same room socially interacting on a recurring basis." In this sense of the word, neither EA nor rationality is a community. I agree that we should not expect Ben/Alice/Chloe to be in the same community with Kat/Emerson, for this narrow sense of community. And my assumption is that they weren't on the day before Ben made his post. And that is fine.

There is a broader sense of the word "community", which we might define as "an extended social network with shared identity and values", which does apply to EA and rationality. I don't see a reason why two people in a legal dispute shouldn't be able to remain in this sort of community.

Why do you think that third parties shouldn't name an accuser? If an accusation is being handled in the court of public opinion, presumably it is because the public has an interest in the truth of the matter, and therefor I would think that any member of the public who has relevant evidence ought to be able to present it. If the accusation depends on the credibility of the accuser, then the identity of the accuser seems like relevant evidence. If anything, I'd think the accused should be particularly hesitant to name the accuser, at least as a strategic matter, for fear of appearing retaliatory. Third parties, not being under that constraint, might be in a better position to name the accuser.