Herrin

Pathos AI | Software Engineering | Full Time | Hybrid in LA or NYC preferred, Possibility of Remote elsewhere in USA | $(160-190)k

Pathos was founded to revolutionize precision medicine in cancer by harnessing the power of machine learning to transform drug development. Pathos aims to accelerate traditional oncology approaches based upon a comprehensive and integrated data-first strategy to discover and develop drugs for precise biological pathways. Pathos is currently pursuing co-development and in-licensing opportunities to build a portfolio of clinical stage oncology assets, applying its technology to find the right patients and indications, resulting in a highly improved probability of success and faster time to market.

I'm hiring for software engineers, with a focus on cloud infrastructure. We'll be developing infrastructure for data analysis and storage, processing and analysis pipelines, and internal applications. Engineering writes code mostly in Python, while our scientists prefer R. We use Google Cloud Platform for our infrastructure. Currently, the team is just me, but we are looking to hire at least 1 more full time engineer.

An ideal candidate will have at least 2 years of experience developing software in a professional environment, with at least 1 year of that working with Google Cloud Platform.

Nice-to-haves are experience with scientific computing, experience with bioinformatics tools and/or biological and medical data, and experience with Python, R, PostgresSQL, and containers. Experience securing data and applications would also be nice.

There's a small amount more about the company at: https://pathos.com/ Please send a resume to steve+hnhiring@pathos.com

He does, starting at the section that begins:

Possibly even worse is the fact that so much academic writing is kept behind vastly more costly paywalls.

I was an experimental physicist, and I'd say that's a reasonable comparison to make.

The time scales are much longer, unfortunately. Instead of taking one or two years to get a company off the ground, it can be more like five for an experiment[1]. I'd attribute this to slower iteration. You end up throwing away or rebuilding a lot of physical things, many of them bespoke, on the way to a working apparatus.

Also, the compensation is much worse for experimental physics. For he first decade of your career, you work as a grad student, then a post-doc, for slightly above poverty wages. This does have the advantage of meaning you work with very motivated people, but you're giving up a lot for that.

[1] The XENON experiments from the article (10, 100, and now 1 Tonne) have been developed over about 15 years, for example

The Screen Actors Guild, and the players unions for most professional sports (NFL, NBA, MLB, etc). They recognize that their members have different levels of talent, and allow the stars to make more, while still ensuring basic rights for every member.

I found this more readable and understandable than the Haskell post, although I can't quite say why. It might simply be the repetition.

I'm really interested in Futhark, though I haven't found a project where it would be make sense to use it. But I feel like it has the same potential to make GPU programming not feel overwhelming the same way Elm did with frontend work for me.

It's a reasonable hypothesis. But the new physics might also lie in why we don't see more than 3 generations of charged leptons or quarks.

Adding masses is a trivial extension, though, by giving them masses in the same way that electrons and other leptons do.

There are a few interesting things that we could get from looking at neutrino masses.

The first has to do more with the nature of the mass than the mass itself. In the standard model, electrons, muons, and taus get their mass from the Higgs field. There's a way for neutrinos to get their mass in other ways, but it requires them to be their own antiparticles. And this gives a satisfactory answer as to why their masses are so tiny, and suggests some new particles (although at electroweak unification scale, so not anything we're going to achieve with a collider any time soon). There are a number of double-beta decay experiments trying to measure the Majorana mass of neutrinos.

The other would be if the neutrino hierarchy is "inverted". The tau is heavier than the muon is heavier than the electron. Right now, with neutrinos, we can only measure the difference in masses. And so it's not clear if the neutrino with the most election portion[1] is the lightest or if the neutrino with the most tau portion is. The latter would be "inverted" from what we expect, and trying to figure out why might be interesting, though I don't know that it immediately implies new physics.

There're also other things related to masses that are interesting. Neutrino oscillations are determined by the differences in mass. Looking at these, we might be able to discover more generations of neutrino (beyond electron, mu, and tau), which would be new physics.

[1] the mass eigenstates of the neutrino (that is, the things with well-defined masses) are not weak force eigenstates, so they contain mixtures of the electron, mu, and tau neutrinos

I remember reading this a decade or two ago, but I never could find such a plot to confirm this. Fortunately, someone generated one here back in 2016: https://hsm.stackexchange.com/questions/264/timeline-of-meas...

If the CDC weren't explicitly prevented from studying gun violence[1], they might be able to look into these problems. It's too soon to know if the recent (2018) changes to the interpretation will make a difference.

[1] https://en.wikipedia.org/wiki/Dickey_Amendment

We're in an investee's market right now. Investors are throwing large amounts of money at anything that might have a positive return. This gives the founders/CEOs/etc. a lot more power to get away with this kind of stuff. If their behavior scares off one investor, there will still be a line waiting to give them money.

When things swing the other way, and not as much money is floating around waiting to be invested, investors will be able to hold the investees to higher standards.

ROOT is my go-to example for peak Object Oriented in the 90s. Look at the inheritance for something like a 2D histogram of doubles[1].

Trying to find the documentation for how to draw an arrow on a plot was always fun, because searching for "TArrow ROOT" would inevitably get results for "taro root".

The XRootD[2] project is pretty interesting, though, and I feel like the software industry is going to have to start dealing with similar data problems before too long.

[1] https://root.cern.ch/doc/master/classTH2D.html [2] http://www.xrootd.org/

This was the first task I ever received and it flickered my curiosity. In the problem set, the dataset contained survey and demographic attributes like race, age, computer proficiency, immigration status….etc.. to predict actual work performance.

I completed the survey which was asking me workplace situational questions testing my unconscious biases and then quickly rejected the dispute.

This sounds like the "consulting company" was someone running a scientific experiment to see who is willing to create racist/ageist/otherwise unethical models.

You have my sympathy. I have been searching for the annotated version for years now. Any readable versions seem to have disappeared.

This is a good reminder for me to make backups of the ebooks I do have.

I would love to see Anathem as a 3 season animated series.

For what purpose? See: https://stressgrid.com/blog/benchmarking_go_vs_node_vs_elixi...

I don't know about the one in Finland, but the one in New Mexico is in the Permian Basin[1]. There are oil and gas resources all over the place. There are also Potash mines right next to the New Mexico facility (and it itself is a salt mine). Leaving it unmarked poses a real risk that someone someone will drill or dig into it accidentally looking for resources.

[1] https://en.wikipedia.org/wiki/Permian_Basin_(North_America)

For anyone interested in things like this, I'd recommend "Deep Time" by Gregory Benford[1]. It has a chapter devoted to WIPP[2] and this messaging.

[1] Amazon link: https://www.amazon.com/Deep-Time-Humanity-Communicates-Mille...

[2] Seems to be online, albeit poorly formatted, here: https://www.physics.uci.edu/~silverma/benford.html

We never had to use it, thankfully.

Our experiment (like the XENON experiments) used liquid xenon. The boiling point is roughly -110 °C, so it requires cooling to stay liquid. If that cooling had failed, the xenon would have started to boil and turn to gas. Gaseous xenon takes up something like 300x more volume than liquid.

So we had a few things to deal with this. We had some pretty giant UPSes to provide backup power. Imagine a shipping container filled with lead acid batteries. That was enough to keep the cooling running for about a day. During that time, we would start recovering the xenon. That would involve running compressors to stuff it back into bottles before power ran out. We also had a limited ability to cryopump the xenon out. Cryopumping involves cooling a gas cylinder (usually with liquid nitrogen) so the gas condenses inside. But that was always limited by the amount of LN we had on hand, which wasn't much.

But suppose we couldn't get all the xenon back into the cylinders before power ran out, or if our compressors failed. First, our detector would have failed. It was made of thin copper to reduce radioactivity. The xenon would have mixed with the HFE (Novec) fluid we were using for coiling. After that, as the xenon continued to boil, it would have burst some burst disks built into the system. And those were connected to the balloon.

The "balloon" was some plastic material, about 10x10x20 meters that we had stashed in an alcove off to the side of our experiment. It would have hopefully contained the xenon.

Since we never had to use it, I'm not completely sure what the process would have been. We would have shipped it off to some industrial gas facility, and they have the equipment to distill it out. And then we'd probably have to spend more time purifying it ourselves.

I doubt such a thing has ever been used. Most experiments, even underground ones, weren't dealing with the constraints we were. The mine we were in wasn't dedicated to science, and a dedicated facility would have had better support. For example, we couldn't run generators to deal with a power failure because there were limits to how much diesel equipment could be running underground with only natural ventilation, and the ventilation fans didn't have backups. Likewise, a dedicated facility would have had a better supply of liquid nitrogen. We only had one portable dewar, rather than a large tank.

I worked on a different xenon-based experiment[1] for my PhD. We were looking for similar rare decays to learn about neutrinos. Our observation of double beta decay of xenon 136 used to be the record for rarest decay.

Xenon has some very nice properties as a particle detector for rare events:

1. For one thing, it's a noble element, and so it can be purified very well to reduce background decays of other elements.

2. It's also a very heavy atom, and so it's self-shielding from external radiation. The core of the detector is shielded very well from radiation coming from the outside, so any signal you see there is most likely from decays of xenon or from things like dark matter WIMPs that don't interact much with matter.

3. It is a natural scintillator[2]. It gives off light when an interaction or decay ionizes the xenon atoms. That lets you actually detect the event, and by collecting the scintillation light, and the electrons from the ionization, you can get a decent measurement of the energy of the event.

4. It's recyclable. The XENON1T experiment follows the XENON100 experiment. The 100 kg from XENON100 were reused in XENON1T, and the tonne from XENON1T will be reused in future experiments. So the cost gets amortized.

$120/g sounds on the expensive side. The price is always changing based on supply and demand. One manufacturer deciding to use xenon in some process, or finding a way to replace xenon with argon, can swing the price by an order of magnitude.

As for leaks, I can say on our experiment we took the possibility very seriously. The entire xenon gas system was made of ultra-high-vacuum plumbing, and we helium leak checked every connection. When the xenon was outside the experiment in bottles, we had sniffers around the bottles to make sure they weren't leaking. We also had emergency systems in place if we needed to recover the xenon, including a "balloon of last resort" that would've captured the xenon in the event of a catastrophic failure.

[1] https://www-project.slac.stanford.edu/exo/

[2] https://en.wikipedia.org/wiki/Scintillator

For the most part, I enjoyed it!

I was using it to get more familiar with Rust, and I feel a lot more comfortable with the language now. I doubt I did everything idiomatically, and especially not at the beginning, but I feel like I can solve problems now. I've been putting my code on github [1] as I go, if anyone wants to see how I did things.

One takeaway for me next year is that there are some common things that pop up on multiple days, and if I actually card about being competitive, I'd want to make generic solutions for them before the competition started. Things like working with 2D and 3D grids, pathfinding, and parsing input. But since I was learning a language, the repetition helped.

Another takeaway was that I should not check the problems immediately when they open, which is late for me. Probably about half the time I would tell myself I'd do some simple stuff, then go to bed and finish in the morning. But then I'd get into it, and stay up too late. I do look forward to not having such a messed up sleep schedule next month.

I'll definitely be looking forward to it next year. This was my first year, so I may go back and do some past years to fill the time until then.

[1] https://github.com/steveherrin/adventofcode2018

23andMe | Software Engineer, Data Applications and Services | Mountain View, CA | Full Time | Onsite

I'm looking for mid-to-senior level engineers to join my team at 23andMe. We're a direct-to-consumer genetic testing company, with over 2 million customers, and with over 85% of those consenting to use their data for research. The team's goal is to make the data we collect accessible and easy-to-use for our scientific, therapeutics, and R&D teams. We currently do this through a combination of web applications, APIs, and some data processing pipelines. This involves machine learning, data visualization, and web development. We end up working on the full stack, with more emphasis on the back-end.

A full description is here: https://www.23andme.com/careers/ooqx5fwa/ , but the requirements are flexible; being able and willing to learn is what I value most.

We're also looking for a tech lead for one of the teams that works closely with mine. Their mission is similar, but with a bit more focus on infrastructure and data pipelines: https://www.23andme.com/careers/ozMN3fwV/

You can apply at the careers page, or email me at steve.herrin+hn@23andme.com if you have questions.