Last week, I gave a talk that felt kind of like a culmination for me. It was a keynote for my alma matter, the English department at the Graduate Center, CUNY, for their annual ESA conference, which was themed on graduate student work, its many manifestations, and the ways that interdisciplinary research speaks to other work the department.

I knew I wanted to talk about my own fears throughout graduate school, of impostor syndrome, which for me manifested most strongly as a fear of public speaking. So I used that as a launch point. Then I went into reading practices in literary studies, using Eve Kosofsky Sedgwick to explore how reparative reading can take what is traditionally seen to be repressive and turns it into a source of sustenance, of generation. From there I moved, perhaps rather abruptly, into my interest in studying transphobia with AI. Then, in probably what was a very disorienting jump for my audience, I dove deeply into the workings of text generation in machine learning tools. I talked about word vectors and plausibility/generalization in LLMs (see Ted Chiang's fascinating article, for example) and brought them into alignment with Trans Studies and the attachment to "norms" and quotidian affects. I ended by posing this congruence between the two, between text generation and Trans Studies, in the tendency and desire for normalization. That's the case I made, anyway. The talk is accessible here.

There are a lot of threads here, and I'm not sure I can bring them fully into conversation. But the ideas are exciting.

The next step will be to prioritize processing and cleaning the datasets. I'm making good progress with the congressional one, but the state one is still in PDF form. After that I will have to find the right hyperparameters for fine-tuning. And I'm on a clock.

It's a big project. We will see what happens.