Luke Stebbing
Tom is no longer hosting these, but EA NYC has an AI subgroup that meets up every month or so.
There’s a Bayesian-adjacent notion of closeness to the truth: observations narrow down the set of possible worlds, and two hypotheses that heavily overlap in the possible are “close”.
But the underlying notion of closeness to the truth is underdetermined. If we were relativistic beings, we’d privilege a different part of the observation set when comparing hypotheses, and Newtonian gravity wouldn’t feel close to the truth, it would feel obviously wrong and be rejected early (or more likely, never considered at all because we aren’t actually logically-omniscient Bayesians).
The best plausible explanation I’ve seen is that Delta’s serial interval might be much shorter, which would mean R is lower than you’d think if you assumed Delta had the same serial interval as older strains. (Roughly speaking, in the time it would take Alpha to infect R individuals, Delta has time to infect R and for each of those individuals to infect another R, leading to R + R^2 infections over the same period.) That makes it easier for behavior changes and increasing population immunity to lower R below 1.
I’ll defer to Blake if he’s done the math, but it does seem worth weighting correlated risks more strongly if they could take out all of MIRI. The inundation zone doesn’t look populated, though, so you’re probably fine.
If you go with Bellingham, will you be avoiding the tsunami inundation zone?
Do you have a source for B.1.1.7 being dominant in Italy/Israel?
Assuming it’s already dominant there, that strongly suggests that it’s infectious enough to have rapidly outcompeted other strains, but that Italy/Israel were able to push down the higher R through some combination of behavioral change and vaccination.
(Note: I can’t find any sources saying B.1.1.7 is dominant in Italy or Israel, and I’d be surprised if that were already the case.)
Is this essentially just giving you leverage in PredictIt?
This process increased my “cash” on PredictIt by $117, but it looks like it will probably pay out around 15⁄14.75*850 − 850 = $15. If I lost my $117 on some other bet, would my PredictIt balance eventually end up negative?
I just donated $5,000 to your fund at the Society of Venturism, as promised.
Like Stephan, I really hope you make your goal.
This concerns me (via STL):
IRS.gov: Automatic Revocation of Exemption Information
The federal tax exemption of this organization was automatically revoked for its failure to file a Form 990-series return or notice for three consecutive years. The information listed below for each organization is historical; it is current as of the organization’s effective date of automatic revocation. The information is not necessarily current as of today’s date. Nor does this automatic revocation necessarily reflect the organization’s tax-exempt or non-exempt status. The organization may have applied to the IRS for recognition of exemption and been recognized by the IRS as tax-exempt after its effective date of automatic revocation. To check whether an organization is currently recognized by the IRS as tax-exempt, call Customer Account Services at (877) 829-5500 (toll-free number).
Do you think your strategy is channeling more money to efficient charities, as opposed to random personal consumption (such as a nice computer, movies, video games, or a personal cryonics policy)?
A more positive approach might work well: donate for fuzzies, but please extrapolate those feelings to many more utilons. I just used this technique to secure far more utilons than I have seen mentioned in this thread, and it seems like it might be the most effective among the LW crowd.
More and more, if I can do anything about it. (Edit since someone didn’t like this comment: That’s a big if. I’m trying to make it smaller.)
I’ll be in Seattle in two weeks, and I’ll take care of it (final three paragraphs).
Kim, I am so sorry about what has happened to you. Reading your post was heartbreaking. Death is a stupid and terrible thing.
Like JGWeissman, I planned to donate $500.
Stephan has been a close friend of mine for the past decade, and when he told me he was planning to donate $5,000, I wrangled a commitment from him to do what I do and donate a significant and permanent percentage of his income to efficient charities. There are many lives to save, and even though you have to do some emotional math to realize how you should be feeling, it’s the right thing to do and it’s vital to act.
He wrangled a commitment from me too: when CI manages a fund for you, I will donate $5,000.
If you’re planning on it, you should get on it now. Cryonics is much more affordable if you don’t have a terminal illness and can cover it with a policy.
People will give more to a single, identifiable person than to an anonymous person or a group.
As a counterpoint to your generalization, JGWeissman has given 82x more to SIAI than he plans to give to this girl if her story checks out.
No matter which study I saw first, the other would be surprising. A 100k trial doesn’t explain away evidence from eight trials totaling 25k. Given that all of these studies are quite large, I’m more concerned about methodological flaws than size.
I have very slightly increased my estimate that aspirin reduces cancer mortality (since the new study showed 7% reduction, and that certainly isn’t evidence against mortality reduction). I have slightly decreased my estimate that the mortality reduction is as strong as concluded by the meta-analysis. I have decreased my estimate that the risk tradeoff will be worth it later in life. I have very slightly increased my estimate that sick people are generally more likely to develop cancer and aspirin is especially good at preventing that kind of cancer, but I mention that only because it’s an amusingly weird explanation.
If this new study is continued with similar results, or even if its data doesn’t show increased reduction when sliced by quartile (4.6, 6.0, 7.4 years), I would significantly lower my estimate of the mortality reduction.
I’ll continue to take low-dose aspirin since my present risk of bleeding death is very low, and if the graphs of cumulative cancer mortality reduction on p34 of the meta-analysis reflect reality, I’ll be banking resistance to cancer toward a time when I’m much more likely to need it. I can’t decide to take low-dose aspirin retroactively.
The meta-analysis you cite is moderately convincing, but only moderately. They had enough different analyses such that some would come out significant by pure chance.
Their selection methodology on p32 appears neutral, so I don’t think they ended up with cherry-picked trials. Once they had their trials, it looks like they drew all conclusions from pooled data, e.g. they did not say “X happened in T1, Y happened in T2, Z happened in T3, therefore X, Y, and Z are true.”
Aspirin was found to have an effect on 15-year-mortality significant only at the .05 level, and aspirin was found not to have a significant effect 20-year-mortality, so take it with a grain of salt.
Can you provide your reference for this? I looked at the meta-analysis and what I assume is the 20-year follow-up of five RCTs (the citations seem to be paywalled), and both mention 20-year reduction in mortality without mentioning 15-year reductions or lack thereof.
Edit: Never mind, I found it, followed immediately by
the effect on post-trial deaths was diluted by a transient increase in risk of vascular death in the aspirin groups during the first year after completion of the trials (75 observed vs 46 expected, OR 1·69, 1·08–2·62, p=0·02), presumably due to withdrawal of trial aspirin.
I’d like to see 20-year numbers for people who maintained the trial (and am baffled that they didn’t randomly select such a subgroup).
Despite the simplicity prior, “this nontrivial program implements the behavior I want” almost always fails to be true because the program contains bugs. Even after you fix the bugs, it still almost always fails to be true because it contains even more bugs.
False scientific theories don’t seem to fail in quite the same ways, and I’m not sure how much of that is due to differences in the search processes versus the search spaces (e.g. physics seems quite different from Python).