rossry

Karma: 636

rossry 27 Nov 2021 2:06 UTC
46 points
on: Omicron Variant Post #1: We’re F***ed, It’s Never Over
I give the WHO kudos for picking omicron instead of nu. (Actually, I’m pretty shocked that they did something this common-sensical, and notice that I am surprised.) I spent Friday morning (= Thursday evening, US time) talking out loud with colleagues about the new nu variant and after like two attempts to clarify what the f—I actually meant, multiple people independently joked that it was so bad we should just skip nu and go to omicron.

If you’ve only ever discussed it in text, you’re underestimating how bad it is to use “nu” as an adjective in verbal conversation.

rossry 14 Nov 2021 4:06 UTC
21 points
on: Sci-Hub sued in India
I don’t want to defend Elsevier’s business practices, but I’m confused by the way some of the numbers are used in the argument against. I understand if the OP is too busy with the legal work right now to respond, but I’d be curious to know whether I’m missing something.
1. India has 30x the financial issues with the academic publishing oligopoly as the US
2. The olipology has such high pricing power and Elsevier has 37% operating profit margin
Take both of these numbers at face value—let’s say that Indian researchers face effective costs 30x higher than a US researcher would (and there’s no developing-countries discount like there is in pharma), and that Elsevier spends 63% of revenues on operations and collects 37% as operating profit.

This means that a hypothetical non-profit publisher operating in the same way would give Indian researchers 18.9x the effective costs that US researchers pay now. Or, if it were operating in the same way and collecting zero revenue, it would need to raise £1.6 billion/year to fund operations. (I’m ignoring second-order effects of lower price on demand and costs.)

I would guess that the “Elsevier, but 0% margin and costs 37% less” would not be tolerable to most critics. (Certainly, it’s far from satisfying to me!) So it seems like the actual goal state is not just “Elsevier is less evil” but also has to include one or more of:
1. Elsevier significantly reduces operating costs
2. Elsevier’s operating costs are supported by people other than institutions+researchers (e.g. by government taxes on all taxpayers)
3. Elsevier restructures costs so that institutions+researchers in developing countries are subsidized by institutions+researchers in developed countries.
I’m curious which of these options are endorsed by Connor_Flexmann and other critics.

(I think that #1 is actually the most promising for reducing costs to researchers worldwide. Finding a way to reduce operating costs by 20% and give half to the publisher as additional margin seems like it should be significantly easier than convincing them to voluntarily give back 20% of profits—though the cost reduction to customers is the same in each case!)

rossry 11 Feb 2019 6:23 UTC
16 points
on: Probability space has 2 metrics
The speculative proposition that humans might only be using one metric rings true and is compellingly presented.

However, I feel a bit clickbaited by the title, which (to me) implies that probability-space has only two metrics (which isn’t true, as the later proposition depends on). Maybe consider changing it to “Probability space has multiple metrics”, to avoid confusion?

rossry 28 Apr 2019 4:40 UTC
15 points
on: Counterspells
Strong approval of the overall goal of the post, but here’s a semantic criticism:
In accordance with the Rationalist tradition that requires everything to have a nerdy sci-fi or fantasy name
I parse this as an in-joke (and appreciate it as such), but I do think that regularly minting new jargon that’s likely to carry substantial, conflicting(!) contextual baggage (not all of it appropriate[1]) is...a bad norm for an epistemic community to have.
I also think that the deeper tradition of jargon-forging (as Eliezer practiced it) involved names that sounded nerdy, but _not_ sci-fi or fantasy -- _cf._ Knowing About Biases Can Hurt People, which uses “Fully General Counterargument” in much the same way(?) as you’re using “Counterspell”. “Fully General Counterargument” is slightly more unwieldy, but apart from that is a better piece of jargon—it’s less loaded and by leaning even harder into the implicit snark that winks at the fact that the purported counter- isn’t working at all, makes it even more clear that no, this is not a useful thing.
[1] To belabor this point, MtG counterspells are fully general (edit: okay, not **fully** general, and see Slider below), and a reasonable fit for the term as you’re using it, but D&D (3.5e, at least) counterspells are based on negating a spell by casting a copy of it, which is not what you mean at all. I don’t actually know what “counterspells” are in WoD/Mage, but the risk that they’re even further afield from your intention should be another strike against using the already-loaded handle.

rossry 17 Jul 2018 10:44 UTC
13 points
on: Probabilistic decision-making as an anxiety-reduction technique
I’m confused; why don’t you just pick black with probability 100%? (Assuming that your utility is 1 if you make the on-further-reflection-correct choice, 0 else.)

This isn’t the same thing as knowing what you’re going to think at the end of your fretting—you still don’t know—but the correct response to uncertainty is not half speed, and just picking the 40%-to-be-right option all of the time is the expectancy-maximizing response to uncertainty.

rossry 13 Jul 2018 14:32 UTC
13 points
in reply to: cousin_it’s comment on: Bayesian Probability is for things that are Space-like Separated from You
It seems pretty easy for such mechanisms to be adapted for maximizing reproduction in some ancestral excitement but maladapted for maximizing your preferences in the modern environment.

I think I agree that your point is generally under-considered, especially by the sort of people who compulsively tear down Chesterton’s fences.

rossry 11 Nov 2017 16:54 UTC
13 points
in reply to: Said Achmiz’s comment on: Frequently Asked Questions for Central Banks Undershooting Their Inflation Target
Eliezer, why on earth are you writing about this? [...] Can’t you just link to some econ-bloggers? There’s plenty of them out there, right? (I seem to recall you even mentioning one, in your previous post…)
It is worth noting that Scott Sumner called this post “probably the best single introduction to the market monetarist way of thinking in the entire blogosphere.”

rossry 8 May 2023 2:07 UTC
12 points
0
on: Long Covid Risks: 2023 Update
(You may well know this, but posting for the benefit of other readers.)

Nirmatrelvir, which is one of two drugs that make up Paxlovid, reduces long covid risk by about 30% for medically diagnosed infections (which means it was serious enough to actually get you to the doctor). An optimist might hope the other drug (which is in the same class, although most commonly used as an adjuvant) is also useful and round this to 50%.

...nirmatrelvir, which is one of the two drugs packaged together to make Paxlovid. I’m going to be an optimistic and assume the second drug was included for good reasons, which make this study underrepresent the usefulness of Paxlovid.

Wikipedia: Nirmatrelvir/ritonavir explains that ritonavir (the other drug) is commonly understood to be playing its role by inhibiting your body’s breaking down nirmatrelvir, leading to higher serum concentrations at trough:

Nirmatrelvir is a SARS-CoV-2 main protease inhibitor while ritonavir is a HIV-1 protease inhibitor and strong CYP3A inhibitor. Nirmatrelvir is responsible for the antiviral activity of the medication against SARS-CoV-2 while ritonavir works by inhibiting the metabolism of nirmatrelvir and thereby strengthening its activity.

Wikipedia: Ritonavir adds some detail on this mechanism and adds that its helpful role in antiretroviral therapy for HIV is also commonly understood to be by inhibiting CYP3A4 (the human enzyme that breaks down many protease inhibitors like nirmatrelvir and ART cocktail components).

I don’t have a head-to-head study of nirmatrelvir vs nirmatrelvir+ritonavir close at hand, but ritonavir is responsible for many-to-most of Paxlovid’s drug-drug interactions, which are commercially negative for the manufacturer (Pfizer). Given that there’s no real reason to add those side effects if the ritonavir weren’t significantly helping, it seems pretty reasonable to update towards “ritonavir improves the efficacy of nirmatrelvir on at least one commercially-useful axis”. That axis is likely not specific efficacy against Long Covid (which I think is not particularly relevant to Pfizer’s approval path or commercialization strategy), though you might hypothesize that it would correlate.

rossry 27 Nov 2021 1:58 UTC
12 points
in reply to: dawangy’s comment on: Omicron Variant Post #1: We’re F***ed, It’s Never Over

Both were sent to the hospital but it is unclear whether this was part of a standard procedure or if they were ill enough to need to go.

Testing positive was sufficient to get them sent to the hospital, and they had mandatory PCR testing every ~3 days; this is no evidence about their symptoms.

(I recently went through HK arrival quarantine—in the same hotel, no less—and researched the operating procedure runbook out of personal interest.)

rossry 14 May 2018 7:42 UTC
12 points
in reply to: fortyeridania’s comment on: Affordance Widths
I think I disagree; “tolerance”, to me, seems to point more towards the special case where {B} is some external events and {X} and {Y} are internal reactions. To talk about social affordances, as the OP does, you’d have to talk about the tolerances of others for {B} done by different people [A-E] -- and you’ve made less obvious the fact that the tolerance of [Q] for {B} done by [A] is different than the tolerance of [Q] for {B} done by [E] -- the entire content of the post.

rossry 14 Nov 2021 3:31 UTC
11 points
in reply to: RyanCarey’s comment on: Sci-Hub sued in India
Elsevier journals allow individual authors to make their own published paper “immediately and permanently free for everyone to read and download” for a fee. In the last Elsevier journal I submitted a paper to, the fee was $2,400.

I think this means that a grant conditioned on open-access publishing would just mean that authors will have to pay the fee if they publish in an Elsevier journal—this makes it more like a tax (paid out of grant money) than a ban. Not sure if that would make it more or less effective, on net, though.

rossry 4 Aug 2019 9:12 UTC
11 points
in reply to: shminux’s comment on: Zeno walks into a bar
My (uninformed) guess is “humor, of the slightly mind-sharpening sort”.

rossry 3 Aug 2019 6:41 UTC
11 points
on: Proposed algorithm to fight anchoring bias
Hypothesis: (formed before running the test below) “clearly too high” will be more of an overestimate than “clearly too low” is an underestimate; most midpoints will be high; this will bias estimates high (especially through repeated self-anchoring to high midpoints).

Anecdata: I tried using this method to estimate the number of gates in Hong Kong International Airport. (I’m currently standing in a line in HKIA, and have flown through it at least a dozen times in my life.)
- 40 ~ 10,000
- 40 ~ 5,020
- 40 ~ 2,530
- 40 ~ 1285
- 40 ~ 662 (maybe 351 is neither obviously too high or too low, so I should take it? or maybe it’s slightly high and so...)
- 40 ~ 351 (195 I don’t have a strong opinion about, so I’ll take it)
So, depending on how I decide to stop, I get estimates around 195 or 351. Apparently the true answer is 90, supporting my hypothesis.

Proposal: Maybe use geometric mean instead of arithmetic mean?

I tried to clear my head and try again with geomean:
- 40 ~ 10,000
- 40 ~ 632 (ehhh, 159 is more likely to high than too low)
- 40 ~ 159 (80 is too low)
- 80 ~ 159 (113)
There’s one suspect step around 159, but my guess here is that being anchored by one higher number (632) rather than five (5020, 2530, 1285, 662, 351) is enough to actually make me think “probably less than 159” rather than “no strong opinion about 195″. Terminating at 159 also is an outperformance of arithmetic mean, but feels a bit more like luck around where the midpoints fell.

rossry 7 Aug 2023 18:52 UTC
10 points
1
on: Monthly Roundup #9: August 2023
Former debater here (around 2009-2011). The discussion about kritik strategies in policy debate has been frustrating to read, (even where I agree with parts of the critique of kritik!). Frankly, I think the kritik-critical bloggers have been following the model of “strongest statement I can make while still being true”, and should be read accordingly. (This makes me sad! They are good people and I wish they would do better!)

Some specific notes, as sub-comments:

(parallel discussion of these at the substack post)

rossry 22 Nov 2021 13:58 UTC
10 points
on: A Bayesian Aggregation Paradox
The framing of this issue that makes the most sense to me is ” $P (E | B \cup C)$ is a function of $P (B) : P (C)$ ”.
When I look at it this way, I disagree with the claim (in “Mennen’s ABC example”) that “[Bayesian updating] is not invariant when we aggregate outcomes”—I think it’s clearer to say the Bayesian updating is not well-defined when we aggregate outcomes.
Additionally, in “Interpreting Bayesian Networks”, the framing seems to make it clearer that the problem is that you used $e_{1, 2} + e_{1, 3}$ for $P (E | B \cup C)$ -- but they’re not the same thing! In essence, you’re taking the sum where you should be taking the average...
With this focus on (mis)calculating $P (E | B \cup C)$ , the issue seems to me more like “a common error in applying Bayesian updates”, rather than a fundamental paradox in Bayesian updating itself. I agree with the takeaway “be careful when grouping together outcomes of a variable”—because grouping exposes one to committing this error—but I’m not sure I’m seeing the thing that makes you describe it as unintuitive?

rossry 14 Nov 2021 15:22 UTC
10 points
in reply to: Vitor’s comment on: Sci-Hub sued in India
Yeah, that all checks out from my publishing experiences with them. (I’ve co-authored one paper for an Elsevier journal and have another out for review with them.)

As I say in my reply to Viliam’s comment nephew to yours: I’m confused by the OP’s choice to present the profit margin figure so prominently in “Why does Sci-Hub exist?”, not discuss the true objections about net-negative spending, and then choose comment guidelines that say “Aim to explain, not persuade”. The margin is striking and persuasive, but (assuming they agree with your model of the world) it isn’t the biggest issue with Elsevier!

I have no particular love in my heart for Elsevier here, but I do care about the common standards of how post-writers argue their claims on LessWrong. If the biggest problem here is actually how the other 63% gets spent, but focusing on that makes a less persuasive case to readers, then I’d at least hope that the author would tag it with something like “Actually, this 37% isn’t my real problem with Elsevier; it’s just the thing that I thought would be easiest to understand.”

(I think your reply actually does an excellent job of this—it lays out a single detailed example, tagged with the evidence you’re basing it on, then you lay out the rest of your objections to Elsevier, tagged with an epistemic status.)

rossry 11 Mar 2020 22:44 UTC
10 points
in reply to: Vaniver’s comment on: Why don’t singularitarians bet on the creation of AGI by buying stocks?
Higher variance is worth avoiding (under standard assumptions), but I for one was surprised by how little additional variance one takes on by allocating, say, 10% of one’s portfolio to a single arbitrary bet. In this comment I ballparked it at maybe an extra 0.5% variance.

That said, allocating one’s entire portfolio this way basically requires a rejection of the standard risk-budget assumptions.

(Disclaimer: I’m a financial professional, but I’m not anyone’s investment advisor, much less yours.)

rossry 3 Nov 2019 13:45 UTC
10 points
in reply to: Viliam’s comment on: AlphaStar: Impressive for RL progress, not for AGI progress
Arimaa is the(?) classic example of a chess-like board game that was designed to be hard for AI (albeit from an age before “AI” mostly meant ML).

From David Wu’s paper on the bot that finally beat top humans in 2015:

Why is Arimaa computer-resistant? We can identify two major obstacles.

The first is that in Arimaa, the per-turn branching factor is extremely large due to the combinatorial possibilities produced by having four steps per turn. Even after identifying equivalent permutations of steps as the same move, on average there are about 17000 legal moves per turn (Haskin, 2006). This is a serious impediment to search.

Obviously, a high branching factor alone doesn’t imply computer-resistance, particularly if the standard of comparison is with human play: high branching factors affect humans as well. However, Arimaa has a property common to many computer-resistant games: that “per amount of branching” the board changes slowly. Indeed, pieces move only one orthogonal step at a time. This makes it possible to effectively plan ahead, cache evaluations of local positions, and visualize patterns of good moves, all things that usually favor human players.

The second obstacle is that Arimaa is frequently quite positional or strategic, as opposed to tactical. Capturing or trading pieces is somewhat more difficult in Arimaa than in, for example, Chess. Moreover, since the elephant cannot be pushed or pulled and can defend any trap, deadlocks between defending elephants are common, giving rise to positions sparse in easy tactical landmarks. Progress in such positions requires good long-term judgement and strategic understanding to guide the gradual maneuvering of pieces, posing a challenge for positional evaluation.

rossry 10 Oct 2019 13:09 UTC
10 points
in reply to: gilch’s comment on: When is pair-programming superior to regular programming?
It’s easy to play armchair statistician and contribute little, but I want to point out that the empirics cited here are effectively just anecdotes. The paper studies 13 pairs and 13 individuals in three assignments in one class at UUtah. Its estimate of relative time costs is only significant to ~ $1 σ$ because development time has variance of (if I backsolved correctly) 65%, which...seems about right. Still, it seems like borderline abuse of frequentist statistics to argue that a two-tailed p<0.05 should be required to reject the hypothesis that pairs finish projects in half the wall-clock time of individuals (which is the null the analysis assumes).

That said, the author correctly identifies that quality matters significantly more than speed. The quality metric, however, is “assignment tests passed” in throwaway academic projects, eliding the questions of what quality failures would or wouldn’t be caught by the review / CI workflows that an industrial project would be going through anyway.

So, finger to the wind, this study feels like it suggests that a pair spends 15% more person-hours (once they get used to each other) before turning their schoolwork in, and do 15% more of the work of the assignment than a student working alone. Consistent with the higher reported work-enjoyment numbers! Definitely a stronger showing than I would have guessed! But definitely not well-abstracted by “no significant result for time; significant improvement for quality”.

What am I missing here?

rossry 12 Mar 2023 15:12 UTC
9 points
6
in reply to: Daniel V’s comment on: “Liquidity” vs “solvency” in bank runs (and some notes on Silicon Valley Bank)
https://www.bloomberg.com/opinion/articles/2023-03-10/startup-bank-had-a-startup-bank-run is the Levine article for anyone else interested in it.

Despite being a near-religious Levine reader, I somehow missed Friday’s post and wrote this post without it. (In my defense, he said on Thursday that he’d be off Friday, then came back to talk about SVB.)

Anyway, Matt has a good phrasing of the unusual weirdness in SVB’s assets, for a bank:

Or, to put it in different crude terms, in traditional banking, you make your money in part by taking credit risk: You get to know your customers, you try to get good at knowing which of them will be able to pay back loans, and then you make loans to those good customers. In the Bank of Startups, in 2021, you couldn’t really make money by taking credit risk: Your customers just didn’t need enough credit to give you the credit risk that you needed to make money on all those deposits. So you had to make your money by taking interest-rate risk: Instead of making loans to risky corporate borrowers, you bought long-term bonds backed by the US government.