All of Kamil Pabis's Comments + Replies

If the treatment is relatively mild, the dropouts are comparable between groups then I am not sure that per protocol will introduce much bias. What do you think? In that case it can be a decent tool for enhancing power, although the results will always be considered "post hoc" and "hypothesis-generating".

From experience I would say that intention-to-treat analysis is the standard in large studies of drugs and supplements, while per protocol is often performed as a secondary analysis. Especially when ITT is marginal and you have to go fishing for some results to report and to justify follow-up research.

The bias introduced is probably usually small, especially when the dropout rate is low. But, in those cases you get very little "enhanced power". You would be better off just not bothering with a per-protocol analysis, as you would get the same result from an ordinary analysis based on which group the person was sorted into originally (control or not). The only situation in which the per-protocol analysis is worth doing is one where it makes a real difference to the statistics, and that is exactly the same situation in which it introduces the risk of introducing bias. So, I think it might just never be worth it: it removes a known problem (due to dropouts, some people in the yoga group didn't do all the yoga), with an unknown problem (the yoga group is post-selected nonrandomly), effecting exactly the same number of participants - so the same scale of problem. In the Yoga context then I would say that if it's really good at curing depression then surely its effect size is going to be big enough to swamp a small number of yoga dropouts. They also only have 32 participants in the trial. I don't know if its a real rule, but I feel like the smaller the dataset the more you should stick to really basic simple measures.
It's a good question. I have the intuition that just a little potential for bias can go a long way toward messing up the estimated effect, so allowing this practice is net negative despite the gains in power. The dropouts might be similar on demographics but not something unmeasured like motivation. My view comes from seeing many failed replications and playing with datasets when available, but I would love to be able to quantify this issue somehow...I would certainly predict that studies where the per protocol finding differs from the ITT will be far less likely to replicate.

The supplement industry and similarly the cosmetics industry is a jarring example of what can happen with no oversight or the wrong kind of oversight. Although, to be charitable to the libertarian position, one can argue that many supplement and cosmetics companies are forced to provide inferior products since efficacious products, even when rather safe, cannot be sold on the free market for various reasons (e.g. higher doses of potassium or retinoids for photoaging).

Thank you for the write-up! Just as a minor quibble, veganism has not been considered the "healthiest choice" ever, or at least not for a long time, if I were to make a guess about "consensus" in the field. While it has been clear for a while that a diet biased towards plants is healthy, the data for the addition of certain food groups (fatty fish, fermented and low-fat dairy, etc) is pretty strong as is the data for the health benefits of individual carninutrients (creatine or even taurine).

As you correctly point out, the issue of residual confounding is ... (read more)

I think the consensus among nutritionists is that a well-planned vegan diet is among the healthiest possible diets. Almost everyone in the US would benefit from "going a bit more vegan". Nevertheless, it is probably not optimal on certain axes.

It would seem that the best diet to improve long-term health is a flexible pescolacto-vegetarian diet supplementing certain carninutrients, e.g. creatine. So not vegan.

Tradeoffs are real and you have to optimize for one thing over another. For example, a standard (unsupplemented) vegan diet may not be optimal for men... (read more)

We have to be careful not to offset the health benefits of a vegan diet. There is a surprising amount of evidence suggesting that low-normal iron stores are beneficial and may reduce cancer incidence, mortality and perhaps increase longevity. Specifically, as strongest, I would point out the FeAst study and numerous recent Mendelian randomization studies on iron and longevity (e.g. Daghlas and Gill 2021). It is prudent to test ferritin to know whether you are too low or too high.

Vitamin D testing certainly could be useful, even though recent clinical trials testing vitamin D supplements are between somewhat to highly disappointing, but deficiency is presumably not strongly linked to veganism.

I do not think this is entirely accurate. Lung cancer in smokers hits unusually young people because, well, they are smokers. Heart disease is a disease of old age and accelerating it somewhat through an unhealthy diet would have complex effects. However, making matters even more complicated, ultraprocessed foods also promote cancers and obesity -- the latter is definitely a huge healthcare burden which does not kill people immediately.

This is hard to model since there can be a shift from a disease that kills slowly to one that kills quickly and early (dem... (read more)

Lung cancer affects old people. Also, while your link claims that lifetime healthcare costs are greater for smokers, it does not claim it is a consensus, but specifically mentions that many people claim the opposite. And that's before getting to Gerald Monroe's point.
4Gerald Monroe1y
I understand the issue wasn't healthcare costs but total cost to the government. It means if you were playing 'simGovernment" from a bird's eye view, and you do not care about the fate of individuals, you are trying to maximize the government's power and revenue, this might be a profitable tradeoff.   You can abstract it as a "card" from an options menu you can select:   [beef subsidy]       + farmer votes       - average lifespan       + healthcare costs       - total cost per citizen       NPV of choice: + <some number> per citizen You can think of this case where as a government official or a gamer playing the "government" as a side your interests aren't aligned with the population's happiness or lifespan as a case of misalignment.

While the term "healthspan" can be useful for public messaging it is not necessary to use it instead of "lifespan" as study after study shows. When the word "lifespan" is used in the correct context people are very willing to embrace even radical lifespan extension. It seems prudent to combine both concepts.

Asked “If doctors developed a pill that enabled you to live forever at your current age, would you take it?” a surprising number of people turned out to be hardcore life extensionists: "There were no differences by age...Among young adults, 40.0% indica... (read more)

Knowing your risk does not change behavior, at least that seems to be the case with genetic risks. That means dietary and lifestyle approaches towards cardiovascular disease are out. As a good approximation, everyone who wants to have a healthy lifestyle already has one*.

On the other hand, it is possible that more people would benefit from wide-spread use of statins and that they could be convinced to actually take them.

Cardiovascular disease is definitely not a neglected cause area. It is a multi-billion dollar industry and a very popular research field. ... (read more)

This is an under-explored topic in modern world-optimization and in Utilitarian theory.  When a large population has a revealed preference for unhealthy lifestyles, do we respect that and include those choices in our total/average welfare calculations, or do we think that overriding their agency is an improvement?
I didn't know these numbers and I didn't know about the Taeuber paradox, but they definitely put Part 5 into perspective.  I wonder if early treatment should be considered a refinement? That is debatable and I honestly don't know the answer. But it does put an upper bound on the benefits of starting early treatment, for which I'm grateful. 

I enjoyed a lot of the other content and hence am now much more inclined to read the EA forums rather than lesswrong. These changes could mean that people like me, who are primarily interested in progress studies and applying science and reasoning to better humanity and themselves, may miss out on relevant AI content when they move to another site. Then again perhaps the EA forums are more relevant to me anyway and I should spend more time reading these.

Nutritionists are not dumb
Let's not be too cynical here. While, yes, nutrition science is short on definite conclusions, it still remains a science. If you want to figure out how to eat healthy, you would find this out the same way you would check whether aspirin prevents cardiovascular disease in certain subgroups or whether paracetamol extends the duration of symptomatic respiratory tract infections.

Step 1: Is there a consensus statement from a reputable professional society? Do different organizations and groups agree? If yes, here is your conclusion. M... (read more)

I don't think you're being cynical enough. Nutritional science is a science, technically, I guess, but it's one of the worst in terms of the quality of evidence it's practical to produce. And there certainy is not as much consensus as you suggest- I don't agree that there's any at all on saturated fats.

I am not sure why we cannot have a vaccine against both strains. The HPV vaccine protects against 9 HPV subtypes, for example. Either I am missing something or it's just the medical establishment moving slowly, as always.

Given the data we have getting an "illicit" fourth booster shot might be the safer play. The mRNA vaccines continue to work, especially against severe disease, the effect is just much diminished.

That's what I did. I'll let you know how it works out. See you on the other side.

Also, is there even any evidence for this assertion? If we stipulate that absolutist monarchies are about as bad as a dictatorship then how did that assertion work out historically? Over the last 10'000 years when lifespans were much shorter dictatorships and related systems flourished. The ascent of democracy has paralleled an increase in lifespans. Correlation does not imply causation, but at least it makes it more likely, whereas the dictator argument is just speculation as far as I can tell.

  1. Thank you for making a great point! Large countries do implement limits on internal migration to ensure political stability. The Chinese system is called hukou if anyone wants to read up on it; I am no expert myself. I would, however, disagree that these limitations suggest there is no free movement. In fact, the very existence of these limitations suggests we should open up compared to the baselines, but perhaps not fully.
    The population of Shanghai grew almost 100% from 14M to 27M within the last 20 years - and the city transformed into a wealthy metropol
... (read more)

You are right, the same is true in Germany as well. There is even some evidence for lower crime rates for certain immigrant groups (e.g., first generation immigrants from Turkey, or SE-Asian/Chinese immigrants, if I recall correctly). Still, more crimes means more crimes, even if this is due to demographics, and the voters will punish the pro-immigrant parties accordingly.

The "lower crime rates for certain immigrant groups" is exactly why these figures are deceptive. If certain groups have lower crime rates, don't lump together high crime and low crime subgroups as "immigrants" and claim that immigrants as a whole are neutral on crime rate.

How limiting are poor corpus quality and limited size for these models? For example, Megatron-Turing NLG was only trained on PubMed extracts but not on PubMed Central, which is part of the Pile dataset. Was this perhaps intentional or an oversight?

Regarding medical texts, I see many shortcomings of the training data. PubMed Central is much smaller at 5 million entries than the whole PubMed corpus at 30 million entries, which seems to be unavailable due to copyright issues. However, perhaps bigger is not better?

Regarding books, how relevant is the limited s... (read more)

Not very, now. Compute & implementation are more limiting. Corpus quality definitely affects how much bang per FLOP you get, at a moderate (neither extreme nor negligible) sort of constant factor. The Pile is better than what OA used, so an otherwise-identical GPT-3 trained on it would be noticeably better. (But this only goes so far and you will have a hard time doing much better than The Pile in terms of cleaner higher-quality text.) Corpus size is unimportant because existing corpuses are enough: the optimal training method is to do 1 epoch, never reusing any data. But you are usually limited by compute, and the compute you have uniquely specifies the model size and n. Even for Nvidia/MS using their supercomputer with 5k+ A100s, that n is a lot smaller than what The Pile etc already contains. (See the part about over/undersampling and how few tokens they trained on compared to the full dataset. GPT-3 did the same thing: oversampled WP and books, but then didn't use more than a fraction of their CC subset etc.) So in other words, PMC vs PM is irrelevant because even if you bothered to get the full PM corpus, they already had more text than they could afford to train on. They don't want to throw out the other data in order to train on mostly PM, so just PMC is fine. (When you have too much data, what you can do to get some value out of it is you can filter it more aggressively to cut it down to just that amount you can afford - but filtering itself is an open research challenge: lots of nasty software engineering, and you may wind up sabotaging your model if you eliminate too much data diversity by mistaking diverse difficult data for bad data & filtering it out. And if you do it right, you're still limited by the first point that data cleaning only buys you so much in constant factors.)

As far as I can tell, if they suspend one of two available mRNA vaccines this is bound to have zero effect on vaccination rates in the young because the other one can fill the gap.

If you completely neglect the effect this will produce on vaccine hesitant population, and consider only these Nordic countries which have plentiful supplies of both mRNA vaccines and over 70% vaccination rate. 

Do you think this is a problem? It appears to me that no development is possible without some tail risk (which we obviously want to minimize wherever possible!). Can we come up with a realistic world in which technologic progress is used for peaceful purposes exclusively and never causes any negative surprises? Or a world that develops with zero tail risk?

2Adam Zerner2y
IMO yes, it is a gigantic problem. I agree that there are tradeoffs where progress implies some amount of tail risk as a consequence. The thing is that I don't think we are navigating those tradeoffs well. The analogy I like to use is that our technological progress is like giving a machine gun to a child. Bad things are bound to happen. To use that analogy, when/if we mature to the level of Competent Adult or something, that would be the time to start playing with machine guns.
1[comment deleted]2y

My personal experience is that it's true hence I would caution against too much rest. Wrist pain is a very vague term, no idea what you have, but I have battled RSI for over half a decade (mostly of the fingers) and at some point it got so bad that I wanted to quit my degree and it felt like even reading a book or newspaper was too painful. By all means, use your voice, contralateral arm or legs and feet to take over some repetetive tasks.

However, you need to use, strengthen and stretch your wrists as well. Targeted massage and strengthening with a physiot... (read more)

Trying to organize my thoughts on progress a bit:

I do not think we lack a "philosophy of progress" as much as the OP.  I would like to argue that progress is real and that there is decent literature on this topic that not enough people read. Moreover, the topic of progress is a good recruitment tool for EA and rationalism. I find it more exciting and powerful than the bleak nihilism offered by atheism, meek criticism of pseudoscience offered by “the skeptics” movements and the vague (but obviously not misguided) appeals to the noble human nature proff... (read more)

I do not think that is true, at least in Europe where hundreds of millions of people use generic drugs every day. In Germany, in the UK and in Austria a doctor will amost always prescribe a generic when available and people will often buy a generic over the counter. While sometimes too conservative, the very reason why we need the FDA, the EMA, et al. is exactly to make sure that generics* work well -- and they almost always do, one cannot use rare counter examples to disprove that. Do we have any hard data to suggest that generics are somehow unsafe?

* i.e. not some Indian blackmarket knock-off that isn't FDA/EMA approved

As I said, the Ranbaxy saga. It shows that even when there's a whistleblower from the manufacturer that tells the FDA that a manufacturer is fraudulent it takes the FDA years to do something about it. The approval consists of believing the company that the studies they provide are true. The FDA doesn't have subpoena power over Indian factories producing generics and it doesn't do independent quality checking beyond believing the companies.

With such highly subjective soft outcomes a lot depends on the way the question is phrased and interpreted (if self-reporting). Thus comparing different populations and studies is almost impossible without really carefully digging into the original publications and even then it is fraught with problems.

If I have a rash post-vaccine and I go to see my GP or pharamcist, am I seeking medical help and is this worrisome? If I get up and later realize that I need to lie down or else I will faint (vasovagal syncope, around one in ten people have some form of need... (read more)

The sad fact is that we do not even understand mice very well. There is this old joke that can be paraphrased like this: if I were a mouse I could be cancer free and live forever, because it is so easy to cure these guys of diseases. As it turns out, however, this is not true. Within my field it was long gospel that caloric restriction (discovered some 100 years ago) can robustly extend mouse lifespan until studies in the last 20 years called this into question.

What the joke gets right is that we understand humans even less than mice. In fact, despite the ... (read more)

Thanks! From what you're saying, empirically we know some more things about mice, but that doesn't mean understand better the details of processes going in them much more than humans.

I guess this goes back to the issue of defining things and what you mean by hallmarks. If you define your hallmarks broadly enough they may include almost anything while being so vague that they are only useful for posters and ads. In the case of vague hallmarks you'd be right, if you fix them you're all good. But even in this extreme case I do expect the number of vague hallmarks to grow a little bit over time as we learn more. In fact, to me they feel incomplete and ill-defined already.

Looking at the classic "The Hallmarks of Aging" paper (first publishe... (read more)

Aubrey's case for trying to focus on the hallmark is that there are a lot less of them then illnesses and it's thus easier to focus on hallmarks then on illnesses.  It seems that this thesis is basically wrong if they are two vague to be individually targeted. 

My reply comes a bit late since I managed to write a long comment without clicking send and only noticed this now. I will address the errors I see in the TL;DR summary from the POV of a semi-professional biogerontologist:

The disease-based approach to aging this seems to favour is useful, but limited. In fact, if you genuinely want to extend both lifespan and healthspan this excessive focus on the disease-based approach would be inefficient because is inconsistent with everything we know about aging. I would go as far as to say that the disease-based approa... (read more)

Actually, they all do include it, but is is subsumed under stem cell aging, loss of cells and reduced regenerative capacity. Also to clarify what I would consider a misunderstanding. Not everything has to fit. There are probably infintely many causes of aging or at least quite a lot. Most of these fall into the rough categories or "hallmarks" we have come up with like reduced stem cell functioning or damage to biomolecules. Many of these causes are not relevant to immediate life extension which is why they can be ignored for now. Other categories or "hallm... (read more)

I thought the idea of the hallmark was that there was one thing that you could fix and if you fix it then you solved the issue. If you solve stem cell aging you don't automatically get more muscle cells.
Good answer, but I disagree with this specifically. I spent a few days reading up on age-related NMJ dysfunction at one point, and my main takeaway is that it's widely studied mainly because it produces cool images. I have not been able to find any evidence at all that NMJ problems cause age-related loss of muscle strength, despite the large amount of research poured into the subject. (If anyone knows of such evidence, I'd love a link to it.) My current best model is that age related NMJ remodelling is just a side effect of other damage/repair, and doesn't have much functional impact other than making the junctions look a bit different when people image them.

Does not make that much intuitive sense to me because there are a lot of random mutations happening. If the first dose first (or first dose only) strategy reduces the size of the whole SARS-CoV-2 viriome, there will be fewer viruses and less genetic variation in total. More infections in total means more genetic diversity. More infections means that a vaccinated person will be exposed to more sources of infection, more virions, more different genomes over time, thus also increasing the likelihood of mutants able to escape the immunologic response.

Does that assume that the amount of mutations (and therefore the risk of an immune escape) is only dependent on the size of the viriome? But isn't it possible that the risk of an immune escape mutation in the 1-dose vaccinated population is much higher than in the rest of the population? I think if that was much higher, it could swamp the benefit of reduced overall chance for dangerous mutations occurring due to the reduced infections from vaccinating more people (as compared to 2 doses). Not sure of any of this, just trying to think through your intuition.