HNHacker News
TopNewBestAskShowJobs

rcthompson

7,341 karma · joined August 31, 2010

http://darwinawardwinner.github.io/resume/ryan_thompson_resume.pdf

[ my public key: https://keybase.io/rcthompson; my proof: https://keybase.io/rcthompson/sigs/DsaLwPFziJbWiFVlQ5zmSsidG9IxdF6IUZfiINIMj0o ]

submissionscomments
rcthompson··on Infant microbes and metabolites point to childhood neurodevelopmental disorders
I haven't read the whole paper yet, but so far it looks like there's no causal analysis, only associative, so anyone speculating about causality based on these results is putting the cart before the horse. By far the more likely explanation for any association of this type is a common cause. For example, take this sentence from the results:

> Infections during early childhood (birth to 5 years) significantly correlated with heightened risks of ADHD or ASD [...]

One plausible cause of such an association is pleiotropy: a genetic variant that influences one's risk for both infection and neurodevelopmental disorders.

Sadly, it is standard practice in the sciences to report "A is associated with B" using phrasings that the general public (and even many scientists) interpret to mean "A causes B". For example, this sentence from the abstract:

> The findings suggest links to immunodysregulation and metabolism, compounded by stress, early-life infection, and antibiotics.

Or this sentence from the results:

> Penicillin use between 1 and 2.5 years resulted in a 1.6-fold (1.2–2.1, 95% CI) higher likelihood of future ASD (p = 0.0030).

None of these are causal statements, but many people would read them as statements that one thing causes another. As much as I complain, though, I honestly don't know how much I can blame the scientific community for accepting "misleading" statements like this. Because it's also possible that regardless of what wording scientists settled on for expressing an association with no implication of causality, the general public would latch on to that phrase and read causality into it.

Edit: I also want to mention that there's absolutely nothing wrong with associative analyses, as long as you don't try to draw conclusions about causality from them.

Edit 2: Reading further into the results, they report a LOT of p-values and CIs with no adjustment for multiple testing, and many of those p-values are barely less than 0.05 and would definitely not be significant with proper multiple testing adjustment (i.e. many of them have a high likelihood of being false positives).

rcthompson··on MTA board votes to approve new $15 toll to drive into Manhattan
> A better fix would be to just make transit more attractive, not make driving less attractive.

I agree, but at least this makes them relatively correct, if not absolutely.

rcthompson··on MTA board votes to approve new $15 toll to drive into Manhattan
For me at least, it will also make driving to Manhattan and parking actually more expensive than taking the train, which it should be. The fact that both options are currently about the same price for me has had me thinking that something somewhere is deeply wrong.
rcthompson··on Bug hunting in Btrfs
I'm pretty sure it's a kernel thread, not a process, since it's part of the filesystem. So it can't be renice'd.
rcthompson··on Bug hunting in Btrfs
Writing to a corrupted filesystem by default is bad design. The corruption could be caused by a hardware problem that is exacerbated by further writes, leading to additional data loss.
rcthompson··on Try to make sudo less vulnerable to Rowhammer attacks
> Dynamically-typed languages similarly should be able to choose enum values at runtime in many cases.

I wonder, is choosing random enum values at runtime more secure against Rowhammer than just having fixed values that were chosen randomly once and compiled in, since presumably the attacking code now has no way to know which bits it needs to flip? If so, it might even be desirable to implement this as a "secure enum" in a compiled language.

rcthompson··on Try to make sudo less vulnerable to Rowhammer attacks
Well, that quote from Wikipedia says the published paper didn't include a working exploit. That might be true even if a working exploit was available after the paper was written and submitted but before it was published. (I don't know if this is the case here, but this sort of thing is common in scientific publishing.)
rcthompson··on Try to make sudo less vulnerable to Rowhammer attacks
Couldn't compilers be configured to use such values for for any enum type? And maybe even auto-insert the appropriate check in the final unchecked else anywhere that enum type is otherwise exhaustively checked?
rcthompson··on I'm sorry but I cannot fulfill this request it goes against OpenAI use policy
What happens when ChatGPT apologizes instead of answering your question about whether the text is an apology or a product description?
rcthompson··on Adventures with compression
Random compression-related story time:

I once had to run a program hundreds of times and capture all the stack traces to help the author catch a stochastic bug. I zipped up all the stack traces and sent the resulting file to the author. The compressed file was only a few dozen MB, and the uncompressed stack traces all together were only a few hundred MB... on my ZFS filesystem with compression enabled. When the author unzipped the file on their end, they were surprised to find around 10 GB of very repetitive stack traces.

rcthompson··on The case for a pipe assignment operator in R
When you do `f(x=5)`, x doesn't get set to 5 anywhere you can see (it happens inside the body of f, which you are treating as a black box). So logically, in the context of the code that's calling f, it's not an assignment operation.

Admittedly this kind of goes out the window with a lot of tidyverse non-standard eval stuff. E.g. if you do `df %>% mutate(x = 5)`, then x does get set to 5 in the data frame, so that is logically an assignment of x in a visible scope using '='. Going by my logic, functions like mutate should ideally use a different symbol for operations that alter the input like this, but we run up against syntactic limitations of the language.

rcthompson··on The case for a pipe assignment operator in R
Yeah, but it's a single-use placeholder. AFAIK It can't do things like:

    x %>% set_names(., str_c("X_", names(.))
All the underscore does is let you pass the piped argument as a named argument instead of the first otherwise-unfilled argument.
rcthompson··on The case for a pipe assignment operator in R
Because in that context the equals sign makes sense, e.g. "Call function F with X equal to 5."
rcthompson··on The case for a pipe assignment operator in R
I use magrittr pipes pretty much exclusively because of the extra features relative to the native pipe, but I also respect the R devs' decision not to add those extra features to the core of the language. The magrittr pipe adds additional syntax beyond the pipe itself, namely the '.' variable to refer to the piped value (essentially it's just Perl's '$_'). It's incredibly useful, but it's not particularly intuitive the way the basic pipe is.
rcthompson··on The case for a pipe assignment operator in R
R's indexing is not only 1-based, it's inclusive on both ends. So indexing from 3 to 5 actually gets you elements 3, 4, and 5. I've programmed a bunch with both R and 0-based languages and I greatly prefer R's way of doing indexing.
rcthompson··on The case for a pipe assignment operator in R
I actually love that R uses '<-' for assignment. Assignment is an inherently asymmetric operation: it takes the RHS and assigns it to the LHS, and the arrow visually indicates this. It feels more correct to have a visually asymmetric operator for it.
rcthompson··on Ask HN: What's your "it's not stupid if it works" story?
Ok, will do!
rcthompson··on Ask HN: What's your "it's not stupid if it works" story?
We will be posting it on a preprint server when we're ready to submit it to journal review, hopefully some time in February (but who knows, publishing timelines are murky at the best of times). The title will be something along the lines of "Reconstructing the temporal development of COVID-19 from sparse longitudinal molecular data", though that's likely to change somewhat.
rcthompson··on Ask HN: What's your "it's not stupid if it works" story?
I'm in the middle of such a story right now. I'm doing research on a data set of COVID-19 hospital patients with multiple blood samples over time from each patient. The obvious thing we want to do with this data is to line up all the samples on a single timeline so we can see how the data changes over the course of COVID-19 from infection to resolution. Unfortunately, as with most infectious diseases, no one knows exactly when they were actually infected, which means we can't just sort them by time since infection and be done with it.

So, we set out to find some way of inferring the timeline from the data itself (RNA-seq and other molecular assays from the blood, in this case). The first thing we tried was to apply some standard methods for "pseudo-time" analysis, but these methods are designed for a different kind of data (single-cell RNA-seq) and turned out not to work on our data: for any given patient, these methods were only slightly better than a coin flip at telling whether Sample 2 should come after Sample 1.

Eventually, we gave up on that and tried to come up with our own method. I can't give the details yet since we're currently in the process of writing the paper, but suffice it to say that the method we landed on was the result of repeatedly applying the principle of "try the stupidest thing that works" at every step: assuming linearity, assuming independence, etc. with no real justification. As an example, we wanted an unbiased estimate of a parameter, and we found one way that consistently overestimated it in simulations and another that consistently underestimated it. So what did we use as our final estimate? Well, the mean of the overestimate and the underestimate, obviously!

All the while I was implementing this method, I was convinced it couldn't possibly work. My boss encouraged me to keep going, and I did. And it's a good thing he did, because this "stupidest possible" method has stood up to every test we've thrown at it. When I first saw the numbers, I was sure I had made an error somewhere, and I went bug hunting. But it works in extensive simulations. It works in in vitro data. It works in our COVID-19 data set. It works in other COVID-19 data sets. It works in data sets for other diseases. All the statisticians we've talked to agree that the results look solid. After slicing and dicing the simulation data, we even have some intuition for why it works (and when it doesn't).

And like I said, now we're preparing to publish it in the next few months. As far as we're aware (and we've done a lot of searching), there's no published method for doing what ours does: taking a bunch of small sample timelines from individual patients and assembling them into one big timeline, so you can analyze your whole data set on one big timeline of disease progression.

rcthompson··on C++ Should Be C++
That's not the point. The point is that "N people can't be wrong" is not a valid argument for any value of N.
rcthompson··on Voyager 1 stops communicating with Earth
MITM, of course, stands for Martian In The Middle.
rcthompson··on Bad NEWS, Emacs
Looking at the commit diff, I don't see anything that necessitates a hard fork. The changes are all in elisp, not compiled C code. It looks like just evaluating the old version of register.el (perhaps with a few compatibility changes) in an Emacs session would revert everything to the old behavior.
rcthompson··on Causal inference as a blind spot of data scientists
Yes, that's true, legal requirements definitely become much more of a factor the closer you get from research to product.
rcthompson··on Causal inference as a blind spot of data scientists
I realize this is nitpicking a minor point in your comment, but I don't agree with your characterization of RCTs in medical research as being primarily constrained by laws and regulations. Any time I've discussed research on human subjects with doctors doing that research, the discussion of what is and is not an acceptable experiment has always been primarily driven by the risks of harm to the people involved in the study. Any time the law comes up, it's usually because the law requires an RCT in a specific setting, as opposed to preventing it (e.g. drug trials). (Of course in the setting of starting a company based on some medical product, the situation may be quite different.)

Biologists, if not data scientists, are used to considering indirect evidence for causality. It's why we sometimes accept studies performed in other organisms as evidence for biology in humans; it's why we sometimes accept research performed on post-mortem human tissue as being representative of the biology of living humans; to name but a few examples. A big part of a compelling high-impact biology (or bioinformatics) paper is often the innovative ways that one comes up to show causality when a direct RCT is not feasible, and papers are frequently rejected because they don't to the follow-up experiments required to show causality.

rcthompson··on Show HN: Whiz – A copilot for your terminal
I think the point of the demo is to start with something dead simple that "everyone" knows and get more complex with each example.
rcthompson··on Unity's Trap
They don't come out and say specifically what they're doing, but:

> I have to accept the new terms to continue paying for it, which we have to do to keep legally using Unity, which we have to do to finish our game – because by this point we’re trapped in their platform to the tune of 5 years’ work.

And:

> The only difference with a released game is our recourse: instead of having to abandon development of our game, we’d just have to abandon the game – never patch, never update. Which sucks for players almost as much as it sucks for devs.

It sounds like they're too deep into developing their current game for switching engines to be feasible, so they're presumably going to agree to the new terms and complete development using Unity, release the game, and hope that Unity's current and future terms don't end up being too onerous for them.

And then never ever touch Unity again after that.

rcthompson··on Mars Keeps Spinning Faster, and Scientists Don't Know Why
It seems weird to me that so many of the "falling apart" simulations described in the intro involve planets crashing into each other. I would expect direct collisions to be an extremely rare "failure mode", even once the system destabilizes, especially for the much smaller inner planets. Much more common, I imagine, would be near misses that fling one or both planets into the sun, out of the system, out of the orbital plane, into extreme elliptical orbits, etc.
rcthompson··on EPA Approved a Chevron Fuel Ingredient That Has a Lifetime Cancer Risk
The point is that the upper bound model is much, much easier to run, since you don't have to do any complicated modeling of how much of the substance people are actually exposed to, you just assume the worst case. If this model already gives you the right answer, it saves you the considerable work of coming up with a realistic exposure estimate (and then the additional work of explaining this estimate in a way that people believe you, etc.).

So it's not a case of running 1 model vs. running 2 models, it's running 1 dead simple model vs. 1 much more difficult and complicated model. If the simple model is good enough, you just want to run that. In this case it wasn't, though.

rcthompson··on EPA Approved a Chevron Fuel Ingredient That Has a Lifetime Cancer Risk
There is absolutely a reason to run a model under such unreasonable assumptions. It gives an upper bound. If you ran that unreasonable model and found that the cancer risk was minimal, then you could say with confidence that under a more reasonable model that doesn't massively overestimate the level of exposure, the risk would not be any higher and would likely be a lot lower. As others have noted though, that's not the result that this model produced, so another more realistic model should have been run.
rcthompson··on Framework Laptop 16 Deep Dive – Memory and Storage
That's what the expansion bay is for (among other things).
← PreviousPage 2 of 34Next →