viking_math

Karma: 178

viking_math 27 Apr 2026 2:46 UTC
1 point
0
in reply to: Steffee’s comment on: The false confidence theorem and Bayesian reasoning
So we might become less certain about its eventual location, but more certain that it won’t fall into the danger zone, which is the part we care about.
Yeah, I think that’s exactly the point?

viking_math 26 Apr 2026 18:39 UTC
2 points
0
in reply to: Steffee’s comment on: The false confidence theorem and Bayesian reasoning
One can argue about whether, in theory, there’s a difference there or if true randomness exists. But I think that it’s irrelevant and that practically speaking, there is a difference. In the case where collision is actually going to happen, more epistemic uncertainty (i.e. worse measurement data) cannot change that fact, but more aleatory variability (i.e. applying random forces to the satellites) can actually make them less likely to collide. Asking whether the “random” forces applied could theoretically be known does not change that fact.
Similarly, in the Knox example I gave above, it made sense for LW readers to say 35% chance of guilt (because they haven’t investigated) and it also made sense for komponsito to say 1/1000 chance (because he had). LW readers were not saying that more than 1 out of 3 times where the evidence looked the way it did, the accused was guilty; they were just expressing the fact that they had not looked at the evidence. Komponsito was saying that the accused is guilty 1 time out of a thousand, because he had looked at it.

viking_math 13 Jan 2026 17:22 UTC
1 point
0
in reply to: JBlack’s comment on: The false confidence theorem and Bayesian reasoning
I think this is essentially the solution mentioned in the paper.

viking_math 12 Jan 2026 21:06 UTC
1 point
0
in reply to: Greg D’s comment on: The false confidence theorem and Bayesian reasoning
Maybe this is obvious to you, but a lot of the content on this site is about explicating common errors of logic and statistics that people might fall for. I think it’s valuable.
Thank you. Maybe I over-indexed on using the satellite example, but I thought it made for a better didactic example in part because it was so obvious. I provided the other examples to point to cases where I thought the error was less clear.
The lesson to draw from the false confidence theorem is “be careful,” not “abandon all the laws of ordinary statistics in favor of an alternative conception of uncertainty.”
This is also true. Like I said (maybe not very clearly), there’s more or less 2 solutions—use non-epistemtic belief to represent uncertainty, or avoid using epistemic uncertainty in probability calculations. (And you might even be able to sort of squeeze the former solution into Bayesian representation by always including “something I haven’t thought of” to include some of your probability mass, which I think is something Eliezer has even suggested. I haven’t thought about this part in detail.)

viking_math 11 Jan 2026 17:07 UTC
1 point
0
in reply to: Richard_Kennaway’s comment on: The false confidence theorem and Bayesian reasoning
I didn’t look for, and so was not aware of, any larger community. I found the 2 linked papers and, once I realized what was going on, recognized the apparent error in a few places. I agree that “decreasing the quality of your data should not make you more confident” is obvious when stated that way, but like with many “obvious” insights, the problem often comes in recognizing it when it comes up. I attempted to point this out to Micheal Weissman in one of the ACX threads (he did a Bayesian analysis of lab leak, similar to Rootclaim’s) and he repeatedly defended arguments of this form even after I pointed out that he was getting reasonably large Bayes Factors based entirely on epistemic uncertainty.
Did you read section 2c of the paper? It seems to be saying something very similar to the point you made about the tracking uncertainty:
for a fixed S/R [relative uncertainty of the closest approach distance] ratio, there is a maximum computable epistemic probability of collision. Whether or not the two satellites are on a collision course, no matter what the data indicate, the analyst will have a minimum confidence that the two satellites will not collide. That minimum confidence is determined purely by the data quality… For example, if the uncertainty in the distance between two satellites at closest approach is ten times the combined size of the two satellites, the analyst will always compute at least a 99.5% confidence that the satellites are safe, even if, in reality, they are not…
So when you say
then you must content yourself with avoiding approaches within around 100 metres, and you will be on the equivalent of the yellow line in that figure.
Is this not essentially what the confidence region approach is doing?

The false confidence theorem and Bayesian reasoning

viking_math10 Jan 2026 17:14 UTC

24 points

19 comments6 min readLW link

viking_math 27 Mar 2024 15:00 UTC
5 points
0
in reply to: quanticle’s comment on: My Interview With Cade Metz on His Reporting About Slate Star Codex
Or explain why the NYT does use the chosen name of other people, like musicians’ stage names.

viking_math 27 Mar 2024 5:31 UTC
1 point
0
in reply to: Mike P’s comment on: On coincidences and Bayesian reasoning, as applied to the origins of COVID-19
Brand new account, reposting old arguments? Not suspicious at all.
Stoyan and Chiu (2024)
“Just because the market was the epicenter doesn’t mean the pandemic started there,” while technically true, is fairly meaningless. If the center were at the lab every lab leak proponent would be shouting at the top of their lungs this conclusively proves the lab leak theory. Debating one particular statistical analysis doesn’t disprove the very elementary technique of “look at the data, it’s obvious” aka https://xkcd.com/2400/.
The multiple spillover theory might be wrong. But then again, so might all of the analyses that Roko cited in his initial post, including the paper about genetic engineering, the Richard Ebright tweet, the RTK estimates, etc. The point of that part was to show that it’s very easy to generate high Bayes factors if you highball favorable pieces of information, ignore unfavorable ones, make convenient assumptions, and multiply numbers together.
https://michaelweissman.substack.com/p/an-inconvenient-probability
This analysis is obviously heavily biased. No Bayes factor at all for the cases being at the market? Again, no LL supporter would seriously say the BF would be one if the cases were clustered near the WIV. This is the exact same sort of highly motivated reasoning that Rootclaim applied, and neither of the judges bought it, for the same reason. The CGG analysis is just wrong, etc.

viking_math 25 Feb 2024 0:38 UTC
1 point
0
in reply to: avturchin’s comment on: On coincidences and Bayesian reasoning, as applied to the origins of COVID-19
They’re not equally unlikely. You haven’t provided any actual evidence for this claim.
Also, why on Earth would we just take the ratio of distances or areas as the probability factor? That’s not how pandemics work.

viking_math 23 Feb 2024 20:31 UTC
1 point
0
in reply to: Richard_Kennaway’s comment on: On coincidences and Bayesian reasoning, as applied to the origins of COVID-19
ICUs were overwhelmed because Covid spread so much. Its hospitalization rate is a few percent and its fatality rate is 1% or so. This is in contrast to diseases like SARS 1 (9.5% fatality rate) or MERS (34% fatality rate). Sure, it’s not mild compared to seasonal flu, but it is much more mild than the obvious things you would compare it to.

viking_math 23 Feb 2024 16:48 UTC
2 points
0
in reply to: avturchin’s comment on: On coincidences and Bayesian reasoning, as applied to the origins of COVID-19
The second thing would be surprising as if the virus can so often jump to humans from animals it will happen closer to its origin in Laos.
Spillover events probably did happen elsewhere, but not all spillover events lead to a pandemic, and covid is usually so mild that it’s not surprising we can’t find any such cases. (I also don’t know if some final important mutation didn’t happen until much closer to the actual pandemic start).
Alternative explanation is following: as the market is one of the most crowded place in the city
This is discussed in the Rootclaim debate. There are many different types of places which served as superspreader events early on, the evidence we have shows the growth rate in the market as the same outside of it, and overall growth didn’t seem to slow down when they closed the market.
If we assume that a worker of WIH was infected at work, this will be completely unspectacular until he started infecting other people. Such person can commute all around the city including to CDC near wet market.
This is also addressed. It would be a fantastic coincidence—much stronger than the one you posited at the start of this thread—if the only place they brought the disease was one of only a handful of other places in the city that a pandemic could actually start. Like, if all the early cases clustered around the WIV, and I said that a HSM worker could have brought it to the lab, would anyone take that seriously?
This, by the way, is exactly the kind of thing that annoys me and which is one of the main issues I made this thread to address. If you make enough favorable assumptions, you can make any hypothesis look good. This is clearly not the best explanation for the available evidence. Merely because you have successfully epicycled your way into a version of the theory which is not obviously impossible doesn’t mean anyone has any reason to think it is even remotely likely. Your arguments aren’t even consistent, as you seem surprised that there were no spillovers between Wuhan and Laos, but then don’t seem at all skeptical of the idea that a sick person would commute all over the city and only bring it to 1 place.
I mean, I could point out that the first non-Wuhan case was in Beijing on December 17th (I think, going off memory here) and that someone could have gotten sick in a different city, and then just hopped on a train and immediately went to the HSM, and the WIV isn’t relevant at all. Is this story convincing? Is there any evidence to support it? Does it feel like I am engaging in truth-seeking, or just throwing shit at a wall and seeing what sticks so I can prop up my pet theory?

viking_math 23 Feb 2024 2:12 UTC
1 point
0
in reply to: romeostevensit’s comment on: On coincidences and Bayesian reasoning, as applied to the origins of COVID-19
What would the disjunctive fallacy be? Failing to account for the fact that P(A or B) >= P(A) and P(B)?

viking_math 22 Feb 2024 18:12 UTC
3 points
0
in reply to: romeostevensit’s comment on: On coincidences and Bayesian reasoning, as applied to the origins of COVID-19
https://arbital.com/p/multiple_stage_fallacy/ ?

viking_math 22 Feb 2024 18:12 UTC
3 points
0
in reply to: RussellThor’s comment on: On coincidences and Bayesian reasoning, as applied to the origins of COVID-19
At one point Miller gave a likelihood against LL by a factor of 1e20 or 1e25, I think during the second debate, on genetic evidence. I don’t think he intended this number to be an actual Bayes factor, but rather to show how easy it is to get a big BF by multiplying many small numbers together (see also https://arbital.com/p/multiple_stage_fallacy/).

viking_math 21 Feb 2024 23:26 UTC
3 points
2
in reply to: habryka’s comment on: Less Wrong automated systems are inadvertently Censoring me
I would like to see what Roko has to say about my post, so now I’m very curious how this works. Is this saying that you get rate-limited if you have at least 7 people downvoting you in the past 20 comments, regardless of how many people upvote you or how many times those 7 people vote? Also, does this count both overall and agreement karma?

viking_math 21 Feb 2024 21:07 UTC
4 points
4
in reply to: avturchin’s comment on: On coincidences and Bayesian reasoning, as applied to the origins of COVID-19
What facility? WIV and HSM are at least 6 miles apart as the crow flies, with a big river between that forces anyone traveling from one to the other to go even further than that.
To override this we need some mental equlibristics (I think of meme here but I don’t want to be rude)
No, you just need stronger evidence. ¹⁄₂₀ isn’t that strong, especially for a complex situation with a high number of possible parameters to check.

viking_math 20 Feb 2024 17:12 UTC
1 point
0
in reply to: Davidmanheim’s comment on: On coincidences and Bayesian reasoning, as applied to the origins of COVID-19
To make sure I understand your point… the “Bayes Factors” I give like 1/ 1 million aren’t meant to be taken literally. Rather they’re to show how easy it is to get a high BF in this case, if you do a very quick analysis that doesn’t account for details. I don’t expect this post, on its own, to convince anyone of the zoonotic origin hypothesis.

On coincidences and Bayesian reasoning, as applied to the origins of COVID-19

viking_math19 Feb 2024 1:14 UTC

62 points

28 comments14 min readLW link

viking_math 17 Feb 2024 6:04 UTC
2 points
0
in reply to: Vaniver’s comment on: Brute Force Manufactured Consensus is Hiding the Crime of the Century
I would describe that as dismissing counter-evidence out of hand; it’s trivially easy to answer the question as stated, even if you don’t believe that particular story. In any event, this seems like arguing over semantics. I think that accusing someone of a being responsible for several million deaths requires quite strong evidence, and that a pretty key component of presenting strong evidence is seriously addressing counter-arguments and counter-evidence. None of Roko’s posts do that.
[It seems to me like you’re arguing he’s making procedural errors instead of just combing to the wrong conclusion / using the wrong numbers, and so I’m focusing on that as the more important point.
Sure. For example, he’s making the exact procedural error you describe in your footnote, by failing to consider how likely the genetic evidence is under the lab leak hypothesis, or if any other cities would look suspicious as the starting location of a pandemic, etc. He’s failing to apply consistent levels of skepticism to sources. But the biggest issue, in my mind, is still just not giving the question the level of consideration it requires. (I’m drafting an actual post so more detailed object-level arguments can go there when I’m done).
This is what numbers are for. Is “1000-1” a lot? Is it tremendous? Who cares about fuzzy words when the number 1000 is right there. (I happen to think 1000-1 is a lot but is not tremendous.)
I’m not sure what the point of arguing about the definition of “tremendous” is. If I had written “a lot” instead of “a tremendous amount” would anything substantial change?

viking_math 15 Feb 2024 23:53 UTC
2 points
1
in reply to: Vaniver’s comment on: Brute Force Manufactured Consensus is Hiding the Crime of the Century
I don’t think that’s counts as tremendous certainty.
“Brute Force Manufactured consensus is hiding the Crime of the Century (emphasis mine). Although the post contains the statement “I believe” it doesn’t really express any other reservations, qualifiers, or uncertainty. It doesn’t present or consider any evidence for the alternatives.
this is really not a tremendous amount either.
It certainly seems like it’s supposed to a lot:
For the love of Bayes! How many times do you have to rerun history for a naturally occurring virus to randomly appear outside the lab that’s studying it at the exact time they are studying it?
If it’s not a lot of evidence, then taking this post at face value, what would one conclude is the probability that covid came from a lab? edit: And if it’s not a lot of evidence, is it ok to accuse someone of mass murder with that amount of evidence?
I think we should treat them as being roughly equally terrible.
Well I think this is pretty wild, but that’s beside the point, as this isn’t what the post actually says:
prosecute what I believe is the crime of the century: a group of scientists who I believe committed the equivalent of a modern holocaust (either deliberately or accidentally) are going to get away with it. For those who are not aware, the death toll of Covid-19 is estimated at between 19 million and 35 million.
It would also be very strange for the post to have a bunch of content which is clearly supposed to be evidence that Covid was, in fact, a lab leak, and not just evidence that Peter Daszak tried to bury evidence if the point is simply that hiding evidence is bad.

viking_math

The false con­fi­dence the­o­rem and Bayesian reasoning

On co­in­ci­dences and Bayesian rea­son­ing, as ap­plied to the ori­gins of COVID-19

The false confidence theorem and Bayesian reasoning

On coincidences and Bayesian reasoning, as applied to the origins of COVID-19