BANNED USER: /comment/484167
>Unban in 0d 12h 38m
stoatherd
No bio...
User ID: 1961
Banned by: @naraburns
stoatherd
1d ago
I think it'd be fair to call their position the null hypothesis from which to argue from
Not in general, and certainly not for culture war issues. That's handing a gigantic unearned advantage to any political positions aligned with LLM biases, which are well documented.
Frontier models are definitely a form of evidence
Yes, in the same way that everything is evidence. But that doesn't make their opinion valuable when arguing stuff here.
If we're arguing over some factual point, e.g. "are tomatoes a vegetable?", it might make sense to invoke an LLM's response (though linking to an actual source is probably better). But when we're arguing over some high-level subjective thing (e.g. "who broke more norms, Biden or Trump?") then going "Claude agrees with me that Trump is worse!" is not useful evidence.
The point of this place is to actually discuss things, make arguments, and defend positions; not to invoke an LLM's opinion as an authority.
And the response you got from claude barely disagreed with MK
-
mkc said that prediction markets were the best predictor.
-
Claude said they are not the best predictor.
If mkc believes prediction markets are very good, that's fine. Calling them the most accurate is a deliberately strong (maximally strong) claim. Words do actually mean things. You don't get to make the strongest form of your claim for convenience, then swap it out for a weaker claim when "disproven".
stoatherd
1d ago
Let me try asking you again, as you've somehow ignored the important part of my comment.
Do you at least agree that, in principle, if I could show an LLM outputting beliefs you believe are wrong, that this would suffice to show that LLMs' high-level opinions shouldn't be cited as evidence on the Motte?
stoatherd
2d ago
I accept your apology for falsely attributing a bunch of claims to me.
I don't disagree much with Claude's take here!
Dude, I know. I actually agree with you on the object level. I'm trying to get you to understand that what LLMs believe should not be presented as evidence on the Motte.
You claimed that prediction markets are the best predictor.
Claude disagrees.
No one said "radical" disagreement. I'm solely pointing out that, according to Claude, you are wrong. (I actually back you over Claude here; that's my point!)
This is how it reads whenever someone else posts a comment, and you respond by going "well Claude says not-X". Claude saying not-X isn't relevant! It's just going "my LLM could beat up your LLM".
Do you at least agree that, in principle, if I could show an LLM outputting beliefs you believe are wrong (which I think I'vce done a couple of times by now), that this would suffice to show that LLMs' high-level opinions shouldn't be cited as evidence on the Motte?
stoatherd
2d ago
Yes, I do actually agree -- AFAICT, proper prediction markets consistently outperform other predictors in this domain.
My goal here is just to get magicalkittycat to acknowledge that Claude saying X shouldn't be used on the Motte as dispositive towards X being true.
I'd hoped that by bringing up an example where mkc and I actually agree on the object level (that prediction markets are very good, actually), this might be enough for mkc to recognise the point.
Unfortunately, this isn't what has happened.
stoatherd
2d ago
Stop lying.
You dismiss the prediction markets as fantasy
Stop lying. Where did I say that?
your own attempt at a defense admits they are really good
Stop lying. Where did I say that?
This is how you act anytime someone points out verifiable facts that go against you
Stop lying. Where did I say that?
Are you confusing me with a different user? Because none of what you've said is in my comment.
I actually think that prediction markets are very good. But my comment is about getting you to acknowledge that LLMs are not a source of evidence. This takes basic reading comprehension to understand, so I get that it's easier to make up a bunch of stuff I didn't say. But could you maybe try being honest for once, instead of just lying?
stoatherd
2d ago
the most accurate forecasting method we have
As you regularly invoke LLMs' opinions as evidence, I went ahead and asked Claude Opus 5.5:
Are prediction markets the most accurate forecasting method we have?
Its answer begins:
Not clearly. Prediction markets are among the best general-purpose forecasting tools, but the evidence doesn't support calling them the most accurate method overall. (... +500 words)
Claude made it clear that the answer varies based on topic, so I went and asked whether prediction markets are the best forecasting method for predicting who will win in a given state. Its answer:
In general, no single method is clearly the most accurate at predicting which party wins a US state. Markets, poll-based models and expert ratings perform about equally well, and averaging them tends to beat relying on any one.
Is this enough to get you to admit that LLMs' opinions aren't actually conclusive evidence?
You do, actually, have to pick one:
a) You were wrong to call prediction markets the most accurate forecasting method.
b) You shouldn't cite LLMs' opinions as evidence.
Imagine falsely accusing someone and then not apologising for it! You're right, that would be embarrassing.
I'm sure you personally didn't do exactly the same thing to me lower down in the thread, where you mistook me for another commenter and threw a bunch of accusations at me:
Then, when I pointed out that I hadn't said any of the things you were accusing me of, what was your response?
Nice apology after accusing a person of throwing a tantrum based on your lack of reading comprehension.
The absolute chutzpah to accuse your political opponents of beclowning themselves with false accusations, and then not apologising for those false accusations, a couple of hours after doing the exact thing yourself. Truly incredible.
More options
Context Copy link