Projects
Welcome to my little corner of the web! Here are the projects and events I’m currently working on + maintaining:
- Verified Neural Network Model Builder
about | github
- Learning Theory and Dynamic Epistemic Logic course
We’re in the process of improving the course, so stay tuned!
NASSLLI 2025 link
- Finishing my Dissertation 🙂
draft | poster
portfolio of my professional experience
researcher profile: cv | google scholar | dblp
contact: cckisby ☆ gmail ♧ com
About Me
I’m Caleb, a mathematician and programmer, and currently trying to finish my PhD at Indiana University. I was a script kiddie who learned to program by using game dev engines. At some point, I was taken by an LMU/Coursera class in mathematical philosophy, and from there became somewhat obsessed with ideas from logic, theoretical CS, and cognitive science. (Other inspiring classes and books just fueled the fire!)
I have been very lucky to be able to pursue a career in this direction—the PhD topic I ended up with is on a mathematical theorem that bridges neural networks (a model inspired by brains) with dynamic epistemic logic (very much of logic/philosophy). And the proof suggests an interesting new approach to AI alignment (which I’m in the process of implementing as a usable program). I’m always happy to talk about these topics, or any other interesting projects & opportunities. 🙂
Just for fun
- What do you think about proof assistants? Yes yes yes yes. Proof assistants (such as Agda and Lean) have changed the game for me. I try to use them whenever I can. They aren’t for everyone (some people find them tedious), but for me they are almost like an accessibility tool that accomodates the peculiarities of my own brain.
- What are your thoughts on OpenAI’s model solving Navier-Stokes? Make sure you’ve read the full story. My understanding is that the final stretch of the proof was a collaborative effort between humans and LLMs. But OpenAI claiming the win is a transparent and scummy PR stunt, and undermines the incredible effort that humans (e.g. Buckmaster and Alpöge) contributed towards the proof.
- Will AI take over everything and leave us jobless? I try not to think about it. I’ll refer you to journalist Cory Doctorow, who I believe is speaking the truth. Basically (1) large language models (LLMs) are surprisingly good at the task of next token prediction, and wow can we take that far, but (2) they aren’t that good at replacing humans. Unfortunately (3) tech companies have an interest in convincing you that they are that good, and therefore justifying that their product is worth the (frankly insane) stock value. And (4) many people are being convinced.
The best thing we can do is to stop believing and sharing their PR story.
- You don’t like LaTeX?!? I experience a lot of friction with LaTeX, even after using it for 10+ years. I’ll use it if I have to. I do prefer visual (wysiwym) alternatives that have a similar vision for typing up math. I love TeXmacs, which is a beautifully designed alternative. (And it’s open source!)