maxander

  Location: Boston, MA
  Remote: Yes
  Willing to Relocate: No
  Technologies: Python (e.g., numpy/scipy, scikit-learn, pytorch), Scala, C#, Haskell, Git, Bazel
  Resume: https://maxalex.github.io/resume.pdf
  Email: w.max.alexander@gmail.com

Master's Degree in Computer Science and seven years of experience in bioinformatics, looking for a development/ML/bioinformatics position in the biotech space. Particular expertise with mass spec data analysis and proteomics, but enthusiastically interested in learning about new application areas.

The odds that they would happen simultaneously if they’re completely unrelated seem astronomically small, certainly?

Both are noted as being related to “ongoing migrations,” though AFAICT not related ones. I would bet there’s a human factor connection- e.g., the day before there was a big meeting where a higher-level management gave multiple ops teams go-ahead on their respective plans, resulting in a multiple potentially breaking changes occurring at the same time.

> They should be trying to grab as much energy as possible, for whatever is it they want to do (as should we) and we should see at least some galaxies going partially dark.

This sort of reasoning was developed in the late 20th century, when economies and populations were rapidly growing everywhere and it seemed like this was just a law of nature. We’re already seeing population growth decline (or reverse) in developed nations, and economic growth has turned out to be much more contingent lately, to say the least. We should probably reconsider the argument that an industrial civilization necessarily develops to building Dyson spheres if the like, when we expect a long-term stable human population of <20B and 2% GDP growth for the foreseeable future.

Beyond that, even, is it really unthinkable that a civilization would decide that at some point it had enough stuff for any conceivable purpose? There are other aspects of the 20th-century cultural mindset at work there, too, I suspect.

The article buries a slightly chilling angle on the phenomenon:

> At a military base in Wyoming, Airmen in charge of launching America’s nuclear arsenal loved to eat acid between shifts. Monitoring the nuclear arsenal is a boring job and, to pass the time, airmen in charge of the nukes would get high.

Don’t get me wrong, I’m generally in favor of people having the opportunity to explore the potential of their own brain chemistry should they choose to. But if there’s any single group of individuals on the planet that particularly ought to be be sober...

I get why everyone is wondering whether this is autogenerated jargon, but I think the basic thrust of this is that machine learning algorithms work better as they use larger and larger vectors for representation. I'm pretty sure is just a well-known thing in ML, although the projects listed here may be pushing vector size farther than most.

So for instance, a typical NLP algorithm (although not GPT-3, IIRC) might represent a word as a 500-float-long vector, which is the same as saying the algorithm considers each word as a point in 500-dimensional space. This turns out to have weirdly useful properties, to the point where directions in this 500-dimensional space start to have semantic correspondences (e.g. [0], still one of the coolest things in ML, IMHO.) You can't do the same trick with a 3D space- the algorithm doesn't have enough to work with when all it knows about a word is three numbers.

Another cool example- in gradient descent, you're constantly trying to find the lowest point in a "fitness landscape"; in a 3D landscape, you might easily find yourself in a "valley" where every direction is worse than you currently are (a local minima), and you won't know where to go. In a 500D landscape, it's unlikely that you'll find yourself in a valley where all 500 available directions lead somewhere worse. So the algorithm will be much less likely to get stuck, and this effect gets more robust the more dimensions you have.

[0] https://colah.github.io/posts/2014-07-NLP-RNNs-Representatio...

What could NK conceivably offer China that would pay for getting into a shooting war with another superpower?

If China does support NK in such a situation, that support would be limited to things which are entirely deniable- material and perhaps some light cyberwarfare. Certainly no extra ICBMs, and nothing that would stop the U.S. from rolling over NK in short order (for all that it certainly increase the cost for the U.S.)

Even assuming the estimate of risk given there is correct, and even ignoring the evidence for persistent damage not infrequently arising from non-lethal cases, it’s really odd to me how cavalier people are about that 0.5% risk of death. That’s 1 in 200, so it’s not huge, sure- but if you gave me the choice between running that risk or taking a shot, I know which I would go for. It would even be too high a price for a night out with my friends.

Like, a 1-in-200 chance of losing all the hopes you had for your life and instead dying an agonizing death in some overcrowded ICU, doesn’t sound like it should be treated lightly.

A PhD isn’t a “now or never” thing either- it’s difficult to imagine that a year or two at Google would make the poster a less desirable candidate.

> This is a particularly sci-fi element. I don't think it's immediately obvious how to separate the universe into "meaningful" computation and the sort of computation a stationary rock is "performing" as its atoms jiggle around, such that one can distinguish at the level of the laws of physics themselves between a rock and a car engine or other simple mechanical device.

It sounds like what you’re asking boils down to “is entropy a real thing, or is it subjective / relative to our purposes and measurements.” And that the Zones essentially work by enforcing an “entropy quota” on things according to where they are in the galaxy- things in the Slow Zones must be very high-entropy (and somehow are forced to become so if they weren’t before) and things in the Transcend can be low-entropy in ways we couldn’t currently imagine.

The kicker is that if their demo examples had been something like that, or e.g. students downloading copies of free lectures, or etc, the RIAA would have much less of a case. No actual code would have had to change.

If you have a tool which skirts (bad, overreaching, probably constitutionally invalid) laws, be smart about how you present them!

Presumably, making it more targeted means they can use more, potentially to the point that the effect on the tumor would correspond to a lethal systemic dose using a course of normal doxorubicin. The patient wouldn’t feel any better while undergoing the therapy in that case, but worse tumors could be successfully treated. The effect/toxicity ratio is the key factor in a lot of these kinds of drugs.

"Algebraic reasoning" in this context is a (AFAIK) sort of fuzzily defined thing, but it's a reference to generalized algebra- of which the "algebra" high-school students encounter is a specialized part. Arithmetical expressions like you describe aren't what is meant there.

I would bet they just used SSDs and made sure that the workload they were giving these servers wouldn’t tend to hit the rewrite limit over that time period. (Even then, I’d bet the expected writes-per-time was the source of the “five year” figure.)

I also don’t believe that a blogging platform like Substack would do anything to actively antagonize a superstar blogger like Scott, who obviously has a significant following of blog-readers independent of them.

Interesting to speculate on the “extremely generous” offer from Substack. They’re obviously willing to devote a decent amount of dev time to making Scott comfortable, and I’m sure they have a good idea of how having SSC on board would promote their platform on its take-over-the-blog-market mission, so it could be generous indeed.

From the results:

“Of 50 patients treated with calcifediol, one required admission to the ICU (2%), while of 26 untreated patients, 13 required admission (50%) p value X2 Fischer test p < 0.001.”

Which sounds like as strong a signal as a study of this scale could hope to show.

> Art receptions are generally a breeze. Tech conferences are the worst.

I and some friends had the funny realization awhile back that the sort of scene where people have colorful hair, tattoos, unusual piercings, etc etc, seems to attract more than its fair share of slightly-face-blind people. Which, on reflection, makes a lot of sense.

I don't think any modern cognitive scientist believes that the statement "language is the stuff of reason," even allowing for poetic flair, is meaningfully true (and I'll leave aside that humans are, obviously, much more than text generation machines.) GPT-3 can generate text but the only context it has is its prompt; when humans generate text they have their complex situation in the world (and perhaps even non-worldly factors, e.g. apprehension of mathematical reality) as context. Fitting the latter kinds of context into AI models is the challenge still facing the path to AGI.

At various points, thinking about what I wear to events where well-worn skinny pants are advantageous, I've thought about hip-pouches as a solution to the pocket problem. I've generally found that this runs into an inevitable quandrary:

- To carry my wallet+phone+keys, it will need to be reasonably sized - To be trusted with my wallet+phone+keys, it will need to be durable- built out of leather, or at least something like heavy canvas - It can't be brightly colored or otherwise goofy, because aesthetics - If you're wearing something on your hip that is dark-colored, made of heavy-duty material, and is roughly holster-sized, it will generally look like a gun holster to casual or distant inspection. That's not only a bad fashion statement, that's plausibly dangerous in some circumstances.

A pouch mounted right above the knee might solve this (I don't think anyone wears their pistol there?) if there's a decent way to wear a pouch that's not hung from the belt.

Hook your thumbs through your belt loops. Same effect, slightly less slouchy, and more panache.

In many COVID19 patients suffering from severe symptoms, the virus has already been cleared out of their system- but the extreme immune response itself (cytokine storm, etc) continues to cause damage, with lethal potential. Interesting parallel.

It would be interesting to see how well this translates between small, long-lived dogs and the very large but significantly shorter-lived dogs- particularly how the differences there compare to those between other mammals.

There’s obvious reason to suspect that the aging process in mice (which occurs over ~3 years) is just a different process than the aging process in humans (which occurs over 80+ years), which would obviously be a big problem for all the research into aging that relies on animal models. If dogs, one species with a very diverse set of phenotypes, could serve as a meta-model for how aging processes differ between short- and long-lived creatures, that would be very useful.

A big part of the "DARPA formula" that doesn't seem to get much treatment here is consistency. DARPA is going to spend a few billion this year, and it spent a few billion last year, and it spent a few billion twenty years ago, and you can feel confident they'll spend a few billion next year. That means that if an academic researcher or similar can position their lab in an area that DARPA likes, they have a good shot at reliable funding for the length of a career. With all the economic uncertainties around research, this is tremendously appealing and will cause a lot of smart people to re-orient their entire research programs around DARPA-friendly subjects [0]. A funding organization that might disappear in a few years, by contrast, will get plenty of grant applications, but won't attract nearly the same level of researcher devotion.

(As a side note, I want to pull out a quote I thought was really nice, hidden a ways in:

"DARPA funds wacky things that go nowhere. DARPA programs have a 5—10% success rate and have included things like jetpacks, earthworm robots, creating fusion with sound waves, spider-man wall climbing, and bomb detecting bees. You can’t cut off just one tail of a distribution.")

[0] Which may, incidentally, go a ways towards explaining why DARPA heads describe their projects as "idea-limited"

I don’t know the details of how durable a uranium warhead core is (and I suspect it’s very highly classified) but they’re pretty precisely machined bits of metal. Even a 3%-yield fizzle should be more than enough to de-rate one. At that point the uranium could be recycled to build a new core, but that’s about it.

I could imagine a wider “targeting” beam, fired so that a satellite overhead of the target site could catch neutrino scatter from the warhead. I don’t know what ridiculous engineering would be required to get sufficient neutrino collection for that to work, on a satellite, but if the budget covered a 50GW neutrino beam it might stretch to that, too.

Since it seems like the upshot of all of this is that Katcher is claiming to have discovered a set of signalling molecules which, when injected, rejuvenate tissue and presumably lengthen life, it runs into one of the most fundamental objections that can be leveled at theory of aging (or cure thereof): of the billions of humans that have existed, how is it that none of them have possessed a mutation which (in this case) endogenously produced the signalling molecules which Katcher is proposing as treatment? In the case of this proposal this sort of objection is particularly strong, since (though the blog post is silent as to what the treatment is) it's given that they are fractions from blood, so they're substances which the body produces in some quantity anyway, and which are sufficiently non-dose-specific that Katcher, with limited resources, could devise a reasonably effective regimen.

Parabiosis is one of the most striking (and just weird) phenomena in aging research, so efforts to understand what's going on with it are liable to bear interesting fruit. I'm just pointing out that there's good reason to doubt that this particular "breakthrough" is the "cure."

I don't have access to the article right now either, but I would like to note to the "this enzyme is going to eat all our stuff!" commenters that, in general, most enzymes don't work very effectively in ambient conditions. The abstract doesn't go into detail into how they acheived the "90% in 10h" metric, but I would assume this was in a precisely controlled solution, incubated at an optimal temperature, with fresh enzyme that hadn't had a chance to start denaturing. You aren't going to be able to spray this on the plastic siding of someone's house and watch it wither away (even if home siding was PET and not, IIRC, vinyl or something.)

I guess I can't rule out at that some recombinant bacterium expressing this enzyme might escape the lab and start causing PET objects to get moldy or develop a patina if left out too long [0]. But that would require that this enzyme somehow benefits the bacterium enough that natural selection wouldn't favor dropping the gene for it.

[0] Hrm, there's a thought - what if your refrigerator needed to be refrigerated?

"An enzyme decomposes 90% of plastic in 10h" is more succinct and readable than the real article title, and it's not wrong. A depolymerase is a kind of enzyme [0].

[0] https://en.wiktionary.org/wiki/depolymerase

The famous Douglas Adams quote is relevant: "Space is big. You just won't believe how vastly, hugely, mind-bogglingly big it is. I mean, you may think it's a long way down the road to the chemist's, but that's just peanuts to space."

The odds of an alien civilization picking up one of our probes, even if we send out thousands of them, is preposterously small. Imagine trying to such an object if it had been left floating at an unknown point in the ocean; now consider how much bigger interstellar space is than Earth's oceans.

There were theoretical reasons to suspect that the Large Hadron Collider, by ramming particles together with sufficient force, could create microscopic black holes with approximately the mass of an atom. This turned out not to occur, but yes, at least in theory black holes can be very small indeed.