When a chatbot gets something wrong, it’s not because it made an error. It’s because on that roll of the dice, it happened to string together a group of words that, when read by a human, represents something false. But it was working entirely as designed. It was supposed to make a sentence & it did.

It’s great to see researchers exploring the possibilities of learning about current extinction from past extinction patterns. That was my initial idea for my PhD, and I have deviated from that pretty hard since then, but I still hope to get on that wagon later in my career!

Thank you! Would love to talk more about it any time!!

13/13 And that's that! Here's the full talk. I'll record a longer, higher quality version later, and post it here with a more detailed rundown. And of course I'll let you all know when the paper is published! Lots I still want to do for this project.

Stay tuned!

youtu.be/aS31GnXh3ec

12/13 I'd love to sit here and say "yes, this easier model provides reasonable trees, keep using it!", but SRFBD is the only way to truly accommodate for this data. I think it's worth the implementation challenge, and that it is the future of fossil phylogenetics. And look at this beautiful tree!

11/13 We need simulations to be sure of which model is more accurate, but the fact that we can identify specific model violations and connect them to biases in our estimates is concerning. We tried a different workaround than usual for FBD specimen, and unfortunately it doesn't seem to have helped!

10/13 This, together with the fact that we are not using all the samples for each of our species, might be related to why we estimate lower fossil sampling rates and higher speciation rates with FBD specimen. This leads to younger estimates of divergence times.

9/13 While the first specimen of Epicyon haydeni is correctly placed as a sampled ancestor of the last, the opposite is true for Epicyon saevus. They're tips, and at the same time! Compare that to the same node in our SRFBD tree and you can see the issue.

8/13 So the model is telling us Archaeocyon is a canine! Very interesting, and something to look into.

We do need to talk about where the FBDS workaround we did (using both the youngest and oldest specimens for a given species) fails. In the MCC tree using both specimens, we see some of this:

7/13 A cool example of where they agree: both models estimate a really low posterior probability (<0.005) of the monophyly of Borophaginae. The monophyly of Borophaginae minus Archaeocyon (the first-diverging genus), however, is pretty likely. So is the monophyly of (Archaeocyon, Caninae).

6/13 There's no good way to summarize budding trees currently, so to compare the trees I made MCC trees from a modified posterior sample, removing the first specimen of each tree. So this is a tree the way you're used to: one tip per species. You can see they're pretty similar (RF-distance: 0.19)