Hello!
And what I'm doing here.
Learning mode
I was a PhD student in Cognitive Science, working on understanding how neural circuits reorganize their activity in response to changing environmental conditions (e.g. where in space rewards are located, what sensory inputs signal rewards). Now, as a postdoc, I find myself hungry to explore artificial neural circuits, their activity patterns during and after training, and how these cash out in the form of a neural network’s “behavioral” outputs. Importantly, I’m interested in how we might be able to infer the mechanisms that drive a particular system’s behavior.
I’m learning all of this from scratch—I don’t have a particularly strong quantitative background—and so I have the strange feeling of going back to being a complete novice. At this specific juncture in life, I think this is the most exciting endeavor I could be spending my time on, and I’m grateful for the rare privilege of having time at my disposal!
What to expect from this newsletter
In this space, I will be posting structured notes
consolidating what I have learned from a particular study or review
synthesizing insights from a variety of studies and taking a stab at anticipating actionable applications and practical limitations of the work I read
reporting the results of toy experiments I run myself with different models.
The studies and reviews I’ll be focusing on will run the gamut from mechanistic interpretability work dissecting the nuts and bolts of neural networks, as well as more philosophical takes on the problem of AI alignment with human-devised values.
Posting frequency
I’m so new at this that it will likely take me time to digest complicated studies and run my own experiments. This means I’ll post pretty sporadically for the time being. If I manage to improve substantially within the next year, and find myself acquiring speed, I’ll revise the posting frequency to challenge myself to a steady rate of posts per month.
If you feel you’d get any value out of sporadic posts, feel free to subscribe at the button below!