trombonechamp

The purpose of the research was to publicly announce that that the research subjects (who did not consent to being studied) messed up. The research consisted of submitting patches specially crafted to make sure the (non-consenting) subjects did indeed mess up.

This is really basic. Here is a harder task from 2015 in monkeys with only 96 implanted electrodes (Utah array): https://youtu.be/lVFWIv5lupo?t=1673

Here is a video from 2014 of someone controlling a quadcopter with their mind with zero implanted electrodes: https://www.youtube.com/watch?v=lsraC04Mm8w

I write musicals (i.e. musical theatre):

https://maxshinnpotential.com/music/empty-box-of-people

https://maxshinnpotential.com/music/if-only

Two examples of times I have used Greek letter variable names: (a) Often times the Greek letter is the standard way of referring to a concept, e.g. "beta" in a linear regression. For these situations, the Greek letter would be used in spoken language to refer to the concept and is more clear than trying to describe the underlying concept in English. (b) If you implement a non-trivial mathematical equation, it is clearer to use the variable names which are found in the published equation than to try to rename them. To understand the equation, people will need to read the published description anyway, and it is easier to map one to the other if you don't change the names. Likewise, if you are implementing your own equation, you will usually use Greek letters in the paper, and for consistency, you should use the same variable names in the code as you do in your paper.

I've been learning a new language for the last few years and without Ankidroid, there's no way I would have made as much progress as I did. Being able to study while away from my computer was really a game changer for me. I build my decks with lots of audio, images, and other fancy features of Anki, but I've never had to worry about whether a feature will be compatible with AnkiDroid because everything in AnkiDroid is seamless.

I left a donation - thanks for all of your effort!

Better link: https://www.nature.com/articles/d41586-020-03348-4

Monkeys do the same thing:

Many non-human animals show exploratory behaviors. It remains unclear whether any possess human-like curiosity. We previously proposed three criteria for applying the term curiosity to animal behavior: (1) the subject is willing to sacrifice reward to obtain information, (2) the information provides no immediate instrumental or strategic benefit, and (3) the amount the subject is willing to pay depends systematically on the amount of information available. In previous work on information-seeking in animals, information generally predicts upcoming rewards, and animals’ decisions may therefore be a byproduct of reinforcement processes. Here we get around this potential confound by taking advantage of macaques’ ability to reason counterfactually (that is, about outcomes that could have occurred had the subject chosen differently). Specifically, macaques sacrificed fluid reward to obtain information about counterfactual outcomes. Moreover, their willingness to pay scaled with the information (Shannon entropy) offered by the counterfactual option. These results demonstrate the existence of human-like curiosity in non-human primates according to our criteria, which circumvent several confounds associated with less stringent criteria.

Paper: https://www.sciencedirect.com/science/article/abs/pii/S00100...

Non-paywalled version: https://www.biorxiv.org/content/biorxiv/early/2018/03/29/291...

He is also known for doing ridiculous things to his body in the name of science...

There was a news article about this in Science two years ago, which a couple such efforts: https://www.sciencemag.org/careers/2018/02/crowdsourcing-goe...

Unfortunately you need to roll that yourself in joblib. But it is automatic in bionic (https://github.com/square/bionic), assuming your workflow can be structured in the way bionic likes.

Which methods are you referring to?

For a related example, type "11:00 EST to UTC" into Google and DuckDuckGo. Google says 15:00 because it interprets "EST" colloquially as "the current time on east coast US". DuckDuckGo says 16:00 because it interprets EST literally as "the time on east coast when it isn't daylight savings time" (compared to ET or EDT). It isn't clear which behavior is more desirable.

The poll respondents did not indicate they would be willing to donate to the cause. They indicated they would be willing to donate to "fix it". There is a very big difference between paying to fix a problem and paying for a stranger to work on a project that may end up fixing the problem one day.

> Not a psychiatrist, but I'm pretty skeptical of the approach of trying to sort things according to a few dimensions. ... I don't see why it should be likely to cleave reality at its joints.

One thing to understand about this is that they're not making up these dimensions from thin air. The dimensions are found in a data-driven manner using one of several techniques, most of which fall under the umbrella of factor analysis[1]. If you're familiar with principal component analysis or singular value decomposition, it's a pretty similar idea: you find the dimensions that explain as much variance as possible.

Then, once you have the dimensions, you try to make sense of them in some way. You may do something called factor rotation, which will change what these dimensions are, but you always have the same information among those five factors. Next, you have to give the factors arbitrary names. The names are usually given for the purpose of making it easy to discuss the factors. (This is the same way that the "Big 5" personality scale [2] was developed.)

For instance, if you had a factor which included the survey questions "Do you have trouble sleeping at night?" "Does it take you a long time to fall asleep?" and "Do you wake up not feeling refreshed?", you might name this an "insomnia" dimension. Really, it's just making it easier to discuss the math and data (though the names do end up having a big impact on how people think about the factors).

You can do this with survey data, brain imaging data, genetic data, etc. There are many different ways to find these. That means you can look at how they line up with each other. There are many theories and structures, but a popular one is about the "p factor" mentioned in the article which boils it down to a single dimension. I haven't read the literature for a few years, though, so I can't comment on current thinking about factor structures beyond what the article discusses.

But in the end, the idea is trying to find statistics which are useful and meaningful, either in the clinic or else for guiding future basic research. So when they say "how many dimensions are there?", what they really mean is "how many dimensions can we find that are useful?". It's not a perfect solution, but it is a good start.

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

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

I highly recommend claws-mail. It is lightweight, very stable, and easily scriptable with Python.

It's a bit tragic that he is remembered for something he himself wasn't very fond of. From his biography:

"Do you know something? I hate the Life game. I've really realized that I hate the damned Life game. ... I'm scared of the following thing happening. I'm scared of it becoming another one of these, 'John Conway, inventor of the celebrated Game of Life...' I told you, every time I turn to a book, a new math book, I look up the sacred name and it says: 'Conway, Game of life, pages 34 to 38.' And that's roughly all it says. And how can I say it---it's not character assassination, exactly, it's quality assassination. I regard Life as trash, and frankly. I mean, it was a real part of my life to have discovered it and so on. And I don't think it should be totally removed. But it seems to be all that I'm known for, among the general public. ... This was never a big deal as far as we were concerned. This was just a recreation, a game we played. Somehow it became a bit more important later on, or at least it did in the eyes of other people. I never thought of it as very important, I just thought of it as a bit of fun. In fact, in a way, I felt ashamed of it. I don't think it counts in the mathematical community, or at least in the serious mathematical community. I don't think any of my Princeton colleagues think this Life game is of any importance. I don't know. In a way it doesn't count for me."

Apparently his favorite work was on the surreal numbers: https://en.wikipedia.org/wiki/Surreal_number

It turns out that people with autism do have difficulty detecting emotions. A classic task in the lab is the "reading the mind in the eyes" task (http://socialintelligence.labinthewild.org/mite/), whereby individuals need to identify the emotion someone is expressing by looking at a picture of their eyes. People with autism consistently show lower scores than healthy controls.

Will improving the ability to detect emotions improve clinical outcomes? We don't know, but their initial evidence looks promising: take a look at the JAMA Pediatrics paper (https://jamanetwork.com/journals/jamapediatrics/fullarticle/...) and the Nature Digital Medicine paper (https://www.nature.com/articles/s41746-018-0035-3), which are both linked to in the article:

Would a therapy like the one you described be better than a technology like this? Maybe. But therapy also very expensive, time consuming, and prohibitive for people who don't have time to take their kid to it every week. We are always looking for new approaches. Even if technology like this doesn't work out, learning why it doesn't work will be helpful for developing future treatments which are better. Remember that this is active research, not an established methodology. (Don't be put off by the "breakthrough therapy" designation mentioned in the article, which is a bureaucratic technical term used by the FDA.)

The paper you linked to is about monitoring, and is not useful by itself without a therapist to interpret it and evaluate it. There is not yet established methodology on how to use this kind of data.

On a related note, there has been a lot of interest recently in real-time feedback for enhanced therapy and treatment. There is a growing literature on neurofeedback in particular, such as reducing depression symptoms by training someone to control their neural activity (e.g. see https://www.youtube.com/watch?v=Xb7bgNo3sUs). Exciting stuff!

I'd love to know more about your opera.

I was pretty excited to see this paper but unfortunately it seems like they have a pretty big multiple comparisons problem. The spatial structure of their brain maps and the really strong correlations scream multiple comparisons. Glancing through their methods briefly to try to figure out why, it looks like they are performing their decoding and leave-one-out cross validation independently for each small group of voxels, which should generate a ton of false positives. This is pretty easy to control for, but they didn't do the key control analysis: scramble the trial identity (category) to create null datasets and then recompute the measures. Unfortunately I bet that these null models would have had similar correlations to what they observed.

Have you considered adding a map of all the museums to your site? I don't want to share my location and I don't necessarily want to look for museums in my present location, but I would love to see what is cataloged and look for museums near future travel destinations.

It works: https://i.imgur.com/9Vtah2o.png

I did a similar analysis with a nearly-identical model but without the assumption of a constant stream of people coming from the left. In my analysis, the best model depends heavily on the assumptions: http://blog.maxshinnpotential.com/2019/03/02/are-buffets-eff...

Some of these links go directly to the retracted paper, some go to the retraction notice (where available), and some go to reports of software bugs which impacted others' results but didn't cause any retractions. This list is of course incomplete, but here are a few, mostly pulled from Retraction Watch:

https://dx.doi.org/10.1038/nbt0308-274

https://dx.doi.org/10.1371/journal.pone.0038234

https://dx.doi.org/10.1371/journal.pcbi.0030158

https://dx.doi.org/10.1126/science.314.5807.1875b

https://dx.doi.org/10.1186/1471-2105-5-80

https://dx.doi.org/10.1038/nm0610-618a

https://dx.doi.org/10.1093/bioinformatics/btx811

https://dx.doi.org/10.1200/jco.2015.62.0294

https://dx.doi.org/10.1038/s41562-018-0507-0

https://dx.doi.org/10.1177/0022146515595817

https://dx.doi.org/10.1021/acs.orglett.9b03216

https://dx.doi.org/10.1073/pnas.1602413113

https://dx.doi.org/10.1093/cje/bet075

https://dx.doi.org/10.1001/jama.2019.11954

https://dx.doi.org/10.1126/science.314.5807.1856

I previously made a big list of papers that were retracted due to software bugs. It was intended to go in a manuscript but I had to cut it out because the conference limited the number of references for the camera-ready version. If anyone is interested I can try to dig up the list again!

You should be able to do this with \write18.

eLife doesn't do that anymore: https://submit.elifesciences.org/html/elife_author_instructi...

TeXmacs is a beautiful piece of software, much better (in my opinion) than TeX. I wrote both my bachelor's and master's thesis in TeXmacs, and used it for years for keeping a research notebook, for homework assignments during undergrad, and for general note taking. I am delighted to hear that it is still being developed. Thank you to the developers for their hard work!

What is the benefit of using a SAT solver directly instead of an SMT solver, other than tweaking for a bit better performance in the SAT solver?

A serious problem is that in 99.9% of things that you study in the brain, there will be no sex difference. However, every researcher records sex information in the data they collect. So even if there isn't a scientific reason to think there might be a sex difference in whatever you are studying, you might as well test it because it is zero extra effort. That means that about 1/20 studies will show an effect with $p<.05$.

Thanks for your work, this looks exciting!

I am curious about the following:

> Since the semantics and design of the language ensure fully determinized execution of any code there is actually no real need to perform logging within a block of code. So, logging is not available in the compute language.

While this may be true for execution, this is not true of data on which the program operates. For example, when processing large amounts of data, how should one keep track of statistics, performance information, data irregularities (which don't cause errors but may be useful to look into further), and other such events which might otherwise be logged?

Similarly, suppose the program performs some operation which depends on the system time which fails at runtime for certain system times. If run in deployed mode, it would restart and run the same thing in debug mode, but the environment would be different. How would such an operation be able to be traced back to the particular conditions which caused the error?