cscheid

For folks who've only read the first book and might be somewhat confused, this specific set of values becomes crystal clear about halfway through the second book.

I can't answer for OP, but I did spend a year commuting a bunch and got through ~60 books or so that way.

I went from having time for 10-15 books a year to having time for ~60 books a year, so yes, you do miss a little. But not ~50 books worth of missing. It's easily worth it.

Then, some 35 years later came the Bayh-Dole act, which however well-meaning it might have been, really provided the incentives for universities to turn into fed money capturing enterprises. The rest is history.

This is exactly right. Just like it's better to think of McDonald's as a real estate company with a food business on the side, these days it's better to think of big state schools as mechanisms for ingesting federal research dollars, with an education business on the side. (Never mind that state schools shouldn't be education businesses _at all_, they should be _public services_)

One of the things you learn just from that series is that anything with this much fluorine jammed into it is just asking for trouble. Case in point, https://www.science.org/content/blog-post/things-i-won-t-wor...

2) I can't look at that screenshot and not hear "I leave message here on service but you do not call"

3) RSA's Adleman was the science advisor for the movie, so I'd guess he snuck in the Asiacrypt poster from the beginning

Still, you just gave me reason to mention one of my favorite "no, that won't work either" paper :) on how Bayes will not save you in the presence of model misspecification. Instead of butchering it any further, I'll just point you to this piece explaining the work, written by the author of the paper himself: http://bactra.org/weblog/601.html

When I said "small dataset", I should have said "small subset of all collected points that are initially fed to the model". The issue isn't that you have to collect the data little by little. The issue is that, once you've given a linear model an input that is outside the distribution which you're hoping to model, nothing about the model can be trusted.

So you collect all the data points, but you don't give them all to the model at once. (I'm describing the RANSAC method here now) You start with a large number of "candidate models" that are all fit with a small number of input points, and then test which of the candidate models predict well the points you have not yet given the model. Then you feed the best of these candidate models only the points which it predicts well, and create a more refined, still outlier-free model. This can be proven to work in the presence of a small number of out-of-distribution points.

Statistics is hard. Like, really really hard. Stuff goes wrong all the time. Please leave it to the experts.

If you're going to do this, please look up the methods behind (for example) robust least squares, outlier detection, L_1 regression, etc. The right way to do this is to start with a small dataset that is with very high probability free of outliers, and slowly grow it by never adding points which have large residuals. (If you've done 3D scan registration and image alignment, this is what RANSAC does.)

The principle is, intuitively, that once an out-of-distribution point gets into your linear model, the model is poisoned forever. You can't trust the model to tell you that the bad points are the outliers. The way this paper does is is irreparably broken, sorry :/

The assignments are not directly available, but my email is easy to find.

One final question/request (for which I'd happily pay for!) : it would be awesome to have a path away from Observable's infra if desired. Say I _really_ want to host a particular notebook locally: is something like that planned? I know this is not a trivial feature since notebooks can call other notebooks, but I'd love to develop stuff on Observable knowing that should the worst happen and it doesn't exist anymore, I can run it all locally on my webpage.