I also enjoyed Outer Worlds.

But also: I enjoyed *Starfield* (well, at least until I did all the wandering I wanted to do and started actually working on the main quest), so I recognize that the bar for me is *very* low.

I feel like Fallout 4 was the point where I came to terms with the fact that I don't actually need these games to be *good* to enjoy them.

I mean, I gave up on that game because I found the stories tiresome regardless of them coming together or not, so take anything I have to say about it with a grain of salt, but also isn't there a "true ending" that does "bring them together"?

I pulled a "say what you will about [blank], but at least it's an ethos" on my students the other day and just moved on knowing none of them would pick up on it.

When it comes to confirmation, what matters for robustness can be boiled down to:

1. Does the hypothesis predict robustness?
2. Do the alternatives predict not robustness?

If you answer yes to both, then you've got confirmation! It's that simple. (2/2)

And you can find the code for our R package on my GitHub. This doesn't have all the pretty graphs, but does all the actual calculating you could want for any of the standard tests.

github.com/coreydethier...

If you had any of these questions, you can now find out!

Sam Fletcher, Nada Mohamed, and myself have put together a Shiny app (severity.shinyapps.io/severity/) that illustrates our work, motivates it with examples, and explains the theoretical backing.

FWIW: the *more* substantive issue is that they take "model report" to refer to "a proposition about what the model entails" (p. 46) while I take it to be a proposition about the model's target.

As such, I think their criticisms simply don't land. But it's possible that I'm mistaken about that.

Ultimately, this doesn't matter much to the argument -- the substantive disagreements are more important.

But it's certainly the kind of thing you'd have hoped that the referees would catch!

*That* assertion is pretty clearly wrong. I make *exactly* the same qualification in the paragraph immediately preceding the one they cite.

I then explicitly say that the qualifications are the same in the footnote attached to said sentence!

What's more notable is their assertion that I qualify my claim that models provide evidence in a way that I don't qualify the claim that experiments do (the image is from page 46 of their paper).

So I asked it to write the intro to an ethics paper in my style, since I've never published anything in ethics.

The thesis it came up with?

All injustice is ultimately epistemic injustice.

So now I know what it is like to be roasted by an LLM. (3/3)

I mention this here because ChatGPT -- without being asked -- described my style using exactly those elements that I *aim* for. That was both flattering and slightly worrying.

It then rewrote the intro to one of my published papers. Boring. (2/3)

People have strong opinions about which *entirely fictional* characters should date.

I mean, I guess that's not in the news, but just agreeing that there are a lot things that people have strong opinions about that are much ... further from relevant than Zionism.

(For those without access to PoS, you can find a pre-print on my website or on the archive: philsci-archive.pitt.edu/23817/)

Also included in the paper are discussions of the connections between classical statistics and epistemology and contrasting views about the goal of statistical theory -- should statisticians be more like engineers or logicians?

The impression that it does stems from a misapplication of classical statistics -- one ruled out, on principled grounds, by Mayo, Fisher, Neyman & Pearson, and every textbook I looked at.

To be clear, you also got roasted by two postdocs in their mid-30s.