Posts

You are a Hittite spy. You report directly to the great king Šuppiluliuma I, who rules from his magnificent capital at Ḫattuša. Recently, the Hittites have been on good terms with their neighbours the Hurrians. Still, you can never be too sure about even the best of allies, and therefore the king has sent you to infilitrate the Hurrian bureauracy so that you can pre-empt any conspiracies brewing within.1 You are so good at your job that you have totally compromised all incoming information channels into the Hurrian government apparatus—you can mess with their weather reports, their intelligence dossiers, their tax records, etc....

I started doing research in the summer before my senior year of high school, the summer of 2019. I’m closing in on 4 years now. The time I have been doing research subsumes the COVID-19 pandemic as well as my entire undergraduate education at Georgetown. I was 16 when I started doing research and I am 20 now. If I have an expected lifespan of ~80 years, then I hope I get to have 60 more years of research left in me....

Unless you have been living under a rock for the last five years, you have definitely (if possibly unknowingly) somehow interacted with a machine learning model that uses the transformer architecture. I have spent a couple months poking at little transformer models like GPT-2 and the 19 million-parameter version of Pythia and yet after working at an interpretability startup for a week I realised that I actually don’t have a great understanding of how a transformer works....

Combinatory Categorial Grammar (CCG) is one of the many, many (far too many) syntactic formalisms posited by linguists in the Chomskyian era. CCG has been outlined in work by Mark Steedman, the most recent guide being his 2001 book which I have to read for a class. Unlike most other syntactic theories, I find the mechanics of CCG very elegant. There is no difference between the distributional categorisation of a lexeme and the argument structure that it has; both are neatly contained in the CCG type system, and the connections to type theory just work out very nicely (with the caveat that this is in my limited understanding of type theory)....

I entered the NAACL conference venue, just a couple blocks away from my place in Seattle, with the absolute lowest of expectations. I have attended two other conferences this year entirely online: ACL and LREC. At both, I had minimal interactions with other human beings beyond “standing” at my poster, and most of that time no one was even there to ask about my work. Hence, zero expectations. The Multimodal ML tutorial....