All of Alex K. Chen's Comments + Replies

FWIW, iron levels are mentioned (by Preston Estep's Mindspan) as being particularly bad for the brain (iron increases dementia rates!). Iron increases ROS in the cell by catalyzing the Fenton reaction. As long as one is not feeling too fatigued from it, I don't think that having low-moderate iron levels is a bad thing 

I have been near-vegan for many years (though I occasionally eat eggs) and my iron/ferritin levels are both well within the normal range. I've posted my bloodwork here - (read more)

Don't forget microplastics/nanoplastics too

The smallest can get through the blood brain barrier and they are this generation's air pollution

Lol remember when I hinted at you to get the booster at Future Forum? [not that I cared that much, but just a #remindme thing that was relevant]

I was the most conscientious person in school (had high grades) and school burned me out so much that I developed a long period of ADHD/burnout after (avoiding all the things that made me unhappy) . The thing that happened to Qiaochu Yuan also happened to me. 

1517 fund (Medici fund) is supportive of these people. It has a discord and is generally very "anti-school". Danielle Strachman and Nick Arnett are very understanding (and approachable) of these people.

Hack Club is also a source of them. I know some - like - who are trying to craft an alternative path.

I helped fund one ( ) to SF at age 16, upon which he immediately dropped out of his University of Washington early entrance program to join a new Thiel Fellow at his new startup.

If you... (read more)

I just read these essays!! They almost read like a new Bible - they give SO many lessons on what's moral, what's important, etc.. and they're way more readable than the average lesswrong essay!

My guess is that rejuvenation of the rest of the body [esp kidney/liver] will maybe give the brain an extra 2 decades  [but this is an imprecise estimate]. 8 years is an important number in longevity b/c it corresponds to "mortality rate doubling time".

who almost certainly have incurred more DNA damage than average

This is unclear - one-shot DNA damage can upregulate Nrf2/genetic repair, and does not recursively damage DNA the way transposons recursively damage them.

2Natália Coelho Mendonça10mo
That’s interesting! Thanks for pointing that out. OTOH, they do get cancer at higher rates than the general population, which is suggestive of more DNA damage and/or worse DNA repair.

I have a HIV-positive friend whose epigenetic age is 6 years younger than his real age of ~32. I wonder if his antivirals helped... (they reduce genomic instability from transposons, though I expect this effect to be stronger in older)

This is golden, thank you! I wonder if healthy adults should take ARVs...

What about in neurons/post-mitotic brain cells?

That's unclear yet. It's not obvious how to revert brain cells without losing important information, which is a potential way this treatment could fail. Hopefully there's a way to do it, or else the brain can survive getting turned into stem cells by parts, but that's a clear obstacle.

Yeah this is from the top-down (and doing things from the top-down requires jumping through hoops). People at Foresight care less about these hoops than people elsewhere(+super-open to ppl from weird backgrounds), and Foresight is WAY higher-impact than most organizations.

Bottom-up: there are SOME people I know who hang out with AI people and who understand them inside from the bottom-up, and this is SOMETHING that can help

Is there any progress on casuality or comprehension or understanding of logic without requiring an enormous amount of compute that makes it seem like solving the problem without actual understanding?

Does the MATH dataset have the worst scaling laws of all these tasks? (and math/logic tasks in general?)

On the caffeine/longevity question => would ought be able to factorize variables used in causal modeling? (eg figure out that caffeine is a mTOR+phosphodiesterase inhibitor and then factorize caffeine's effects on longevity through mTOR/phosphodiesterase)? This could be used to make estimates for drugs even if there are no direct studies on the relationship between {drug, longevity}

[aka inverse-compositionality]

Also would ought be able to compare the dose/response curves used on animals and then "translate" them to humans? (eg effects of a study on rats... (read more)

Thanks for the long list of research questions! Yes - causal reasoning is a clear case where decomposition seems promising. For example: We'd need to be careful about all the usual ways causal reasoning can go wrong by ignoring confounders etc

BTW for anyone who knows me, this post is probably my favorite article on the ENTIRE Internet for reasons somewhat similar to my Thiel obsession

You're better off spending the money on empagliflozin/rapamycin/metformin.


Among invertebrates, birds and mammals, experimental paradigms that limit reproductive investment also cause lifespan extension [232]. Hypothetically, the need for repairing and preventing damage to the germline dominates resource allocation strategies, while the somatic tissues age and deteriorate [112]. In support of such theories, modulations of reproduction that eliminate germ cells in C. elegans and D. melanogaster provide effective mechanisms for extending lifespan [232-234], phenotypes that may be caused by heightened resource availability and

... (read more)

Have you thought of to reduce labor costs?

You know, I barely checked LW from 2015 to 2020, and now I check it like, every time I feel like I need some novelty refresh, almost as much as I do Twitter... It has definitely improved since a few years ago

The dot-com crash was also preceded by an extremely obvious and unique bubble that has not been seen since - diversifying/rebalancing during a massive/obvious bubble doesn't take that much special skill or awareness, and we're more aware of bubble dynamics now than 2000.

Throughout the last decade (or last 15 years, really), FAANG stocks (and QQQ) have consistently overperformed the market/index funds, with roughly comparable maximum drawdowns relative to even the S&P. It was clear to many of us technophilic early adopters even in the late 2000s that Amazon/Google were going to take over the world (though I'd replace Netflix with NVIDIA as NVIDIA is just more innovative), and their returns have massively outperformed the market, with much smaller drawdowns. COVID only accelerated the returns from FAANG - however - with... (read more)

3mako yass2y
Have you considered not investing in those things, for that reason? As I understand it, accelerating AGI timelines would be the dumbest thing I could ever do. (Money isn't worth winning if it comes at the expense of imperiling my species (and can't then be used to proportionately de-emperil them)) (My understanding is that alignment work likely requires serial contributions from slow human theorists, can't really be accelerated very much with availability of better hardware. Better hardware moves the deadline back more than it speeds up the alignment theorists, it burns more time than it grants. More powerful hardware is also more amenable to brute force methods, making it more likely that AGI will be realized with very opaque designs rather than meticulously crafted, explicit understandable designs that we might have been able to audit and verify.)
1Alex K. Chen2y [] (and this was pre-covid)
True but you conveniently cut off the dot com crash, in which the NASDAQ QQQ crashed from 107 to 23 (79%) while the S&P500 only fell about 43%. In finance beware the selective "well chosen" example. From the 2000 top that is a return of only about 5% per annum compound.

Does knowing the structure of a protein help with simulating how it responds to any arbitrary/unknown protein/molecule/agonist/antagonist/superagonist? [it seems that even with all the protein structures that we do know well, that finding appropriate agonists of the protein with the desired action is still a huge unsolved problem]. Is simulation a much more difficult problem than "folding"?

This allows us to design "efficient" proteins (proteins designed "intelligently" often do tend to be smaller, less "messy" and "bulky" than naturally-evolved proteins [w... (read more)

Is there an online way to better tag which studies are suspect and which ones aren't - for the sake of everyone else who reads after?

Check out PubPeer.
2Alexey Lapitsky2y
I am using [] with a plugin for browsers, but I would love a similar service with user-generated flags.

So, two years ago I quit my monetarily-lucrative job as a data scientist and have mostly focused on acquiring knowledge since then. I can worry about money if and when I know what to do with it.

Also this knowledge only matters if you do something useful with that knowledge, which I'm convinced that you are, for instance. many other people are not able to create useful knowledge and thus may be better suited for earning2give.

Do you think that applying black box models can result in "progress"? Say, molecular modeling/docking or climate modeling or whole-cell modeling or certain finite-element models? [climate models kind of work with finite element analysis but most people who run them don't understand all the precise elements used in the finite element analysis or COMSOL]? It always seems that there are many many more people who run the models than there are people who develop the models, and the many people who run the models (some of whom are students) are often not as know... (read more)

I would not usually call finite element models black box. It's true that they can be used like mysterious black boxes by particular users, but the gears inside the model do map directly to gears in the physical world, and the people who wrote the code do understand that mapping.

>In line with John’s argument here, we should develop a robust gears-level understanding of scientific funding and organization before assuming that more power or more money can’t help.

How about a metaculus/prediction market for scientific advances given an investment in X person or project? (where people put stake into the success of a person or project?) is this susceptible to bad incentives?

I think the greater concern is that it's hard to measure. And yes, you could imagine that owning shares against, say, the efficacy of a vaccine being above a certain level could be read as an incentive to sabotage the effort to develop it.

in the space of aging (or models in bioscience research in general), you should contact Alexey Guzey and Jose Ricon and Michael Nielsen and Adam Marblestone and Laura Deming. You'd particularly click with some of these people, and many of them recognize the low number of independent thinkers in the area.

I think you have a kind of thinking that almost everyone else in aging I know seems to lack (If I showed your writing to most aging researchers, they'd most likely glare over what you wrote), so writing a good way to, say, put a physical principles framewor... (read more)

More on lipid oxidation: it also depends on the composition (polyunsaturated to saturated fat ratio) of the lipids that compose the cell's walls. You also need to account for all the other highly reactive cell oxidation byproducts (eg 9-HNE, michael adducts, methylglyoxal, CML) as well as the overall redox potential of the cell (eg glutathionine seems to be an abundant antioxidant that reacts with many highly reactive byproducts..)

Additionally in the case of arteriosclerosis, it also depends on the ratio/supply of oxidized cholesterol (particularly keto-ch... (read more)

538 totally outperformed in 2012 on intrade - it seems like there were whales pushing up the romney price on intrade.

Is it even possible to map out "root causes" in a complex system (eg maybe Granger causality in neural networks) when the "cause" could be multiple factors that are jointly necessary - none of them sufficient enough to cause the irreversible feedback loop in itself?

"A prototypical example here would be an abstraction-based decision theory. There, the notion of "success" would not be "system achieves the maximum amount of utility", but rather "system abstracts into a utility-maximizing agent". The system's "choices" will be used both to maximize utility and to make sure the abstraction holds. The "supporting infrastructure" part - i.e. making sure the abstraction holds - is what would handle things like e.g. acting as though the agent is deciding for simulations of itself (see the link for more explanation of that)."

&... (read more)

Interesting analogy, I hadn't thought of that.

This points toward a more general class of questions: when, and to what extent, does it all add up to normality? We learned the high-level ideal gas laws long before we learned the low-level molecular theory, but we knew the low-level had to at least be consistent with that high-level structure. What low-level structures did that constraint exclude? More generally: to what extent does our knowledge of the high-level model structure constrain the possible low-level structures?

One good class of structure for these sorts of questions is causal structure: to w

... (read more)
Problem is, we didn't know beforehand (i.e. in 1800) that the high-level things we saw (like temperatures, heat flow, etc) had anything to do with statistical averages. One could imagine an alternative universe running on different physics, where heat really is a fluid and yet macroscopically it behaves a lot like heat in our universe. If we imagine all the difference ways things could have turned out to work, given only what we knew in 1800, where does that leave us? What low-level structure is implied by the high-level structure?

So like, can you use morphisms to map paths described in one graph to paths described in another graph even if the nodes are different or loosely defined? (eg a functor from one graph to another that creates paths all the nodes that are tagged as "high probability" or all the nodes that have "connectivity matrix exceeding X" to a second graph that is very different from the first graph but which has nodes that still can be ordered by connectivity and have connectivity values that may exceed X?) Where X may be a fixed number or a number that scales accordin... (read more)

doesn't pareto-optimal imply lack of convexity/concavity?

Nope. Roughly speaking, pareto-optimality tells us about a gradient, while concavity/convexity tell us about curvature. That said, if we can randomize decisions, then pareto-optimality under expectations implies concavity: if our frontier is convex, we can take a random mix of two points on the frontier in order to get a point whose expected value is a pareto improvement over the frontier.

Structural genes like the extremely long-lived proteins in nuclear pore complexes don't turn over (similarly, damage to nuclear histone proteins is very difficult to repair). Even small changes in these genes can affect the ability of mRNA and all of the spliceosome proteins to be properly assembled where they're most needed => this gradually sums up to a corrosion of cellular information

They do turn over when the cell turns over, which for most cell types is still way faster than the timescale of aging. They could be a plausible root cause in very long-lived cell types, but I would guess that in long-lived cells they usually do turn over on a timescale faster than decades. This paper [], for instance, finds that nuclear pore turnover is slower than turnover of rat kidney cells, but rat kidney cells turn over in weeks IIRC. NPC could turn over in years, and that would still be fast compared to aging.

Shouldn't smart contracts with staking also allow you to more readily enter contracts where payoffs are unknown? (eg you're not sure if investing in a person or decision will result in the payoffs you want - there's rather a distribution/ambiguity of outcomes). You mention rebalancing - this is where formalized smart contracts allow you to rebalance contracts based on another element of risk if you notice that you've staked too much on options that are volatile in response to investments that have too much time-correlated X1 in them?

You might even be unsur... (read more)

Damage/dysregulation to the control sites are more central to the network - repair genes/proteins like OGG1/ERCC1 or the upstream control factors of everything or kinases. For whatever reason, expression of most repair genes (and heat shock proteins) goes down with time.

Spliceosomes are esp impt too, as are the upstream genes behind lysosome synthesis ( and proteaosome synthesis.

Damage to structural components (like extremely long lived proteins) are harder to repair and simultaneously make it harder for repair proteins t... (read more)

Nope, they turn over too quickly. You'd have to damage most copies at the same time in order for it to have a permanent effect; otherwise the remaining copies will bring us back to equilibrium. (And even if most copies were damaged at the same time, the whole cell should still turn over, so that would also need to be prevented somehow in order to prevent reequilibration.) If expression is decreasing on a timescale of decades, then something upstream must be changing the equilibrium expression level.

Well, the root cause is ultimately the accumulation of small kinds of damage and dislocation (like oxidative/carbonylated damage on proteins/DNA or increase of clogged proteasomes/lysosomes or inappropriate DNA adducts) that ultimately do not get corrected. An oxidative damage event in itself is nothing, but when you combine all of the events integrated in a lifetime, amounts of something.

Sure, but the vast majority of damage types are repaired (in the case of DNA) or removed (e.g. when a protein or cell turns over). So the question is which specific damage types are accumulating. Many kinds of damage increase in count with age, but the vast majority of them turn over too quickly to be a plausible root cause.

But it's pretty suspicious if two different causes both increase the risk of every single one of those things - if we had a complete graph with random weights on all the connections, then some factors should push some diseases up, while others should push down. Instead, we see a variety of mutations/interventions (like progerias, calorie restriction, etc) which all push most of these things in the same direction, which pretty strongly suggests that they're operating through the same pathway.

It's all about the damage to repair/replacement balance. Birds hav... (read more)

>Usually it will copy into non-coding DNA, and then be suppressed, so there's no noticeable effect. But over time, the transposon count increases, the suppressor count doesn't increase, and eventually the transposons get out of control.  

Wouldn't it expand the size of the genome and potentially affect the distance between promoters/enhancers and target genes, causing a loss in a cell's ability to appropriately regulate translation in response to perturbation?

I know some people (like genesis lung) who actively take lysine or antiretrovirals to suppress transposon activity - antiretrovirals may be aassociated with longevity./

To some extent, though presumably the vast majority of copies will be into non-functional sequence, and copies into functional sequence will often result in a defective cell which is quickly removed. The expansion of the genome size shouldn't be significant until the count is already way out of control; a transposon is tiny compared to the whole genome.

I should add that much of the most encouraging progress is in the area of xenotransplantation/stem cell transplantation, artificial organs, and neuronal replacement therapy (it's already being done for Parkinson's, though regeneration of the basal ganglia may be easier than regeneration of the entire brain - the lab is particularly good to follow on this). You don't need an entire mimic of the brain's memory or identity to maintain the continuity of consciousness. (read more)

Uh, I think loss of proteostasis and increased damage to proteins/lipids can be implicated in all types of age-disease (you could theoretically have perfect genome integrity and loss of proteostasis and aging would still occur, though at some pt the loss of proteaostasis would hit the genome.) Similarly, you can have an organism age without inflammation (think of single-celled organisms), telomere damage, oxidative stress (though oxidative damage is one of the most common forms of damage), or senescence (all of these are just accelerants). More complex org... (read more)

Plausible as a common intermediate cause, but not as a root cause. The proteome generally turns over on fast timescales, so it's in equilibrium on fast timescales. If it's changing on a timescale of decades, then something other than (the fast-turnover parts of) the proteome must be causing that change.
1Alex K. Chen2y
[on the stem cells - [] ]

The chances of climate change making Phoenix uninhabitable >>> the changes of being cryonically revived. Keep in mind that the energy required for AC increases as the square of temperature difference between inside and outside, and very few people really know how to deal with temperatures that regularly go above 120F, which could very well happen to Phoenix in 60 years.

Load More