Also thanks to
@michael for introducing me to Kush in the first place!
Wow. I have a comment out in Nature Machine Intelligence about why people don't do data work (enough) and what we might do to encourage it.
https://www.nature.com/articles/s42256-023-00673-x.epdf
Many thanks to everyone at IBM who supported my somewhat circuitous internship: Kush Varshney for letting me explore and being the most encouraging mentor I've ever had. Payel Das for truly making the paper happen at every step. Prasanna Sattigiri, Inkit Padhi, and Pierre Dognin for their guidance and contributions along the way.
The incentive gap in data work in the era of large models | Nature Machine Intelligence
There are repeated calls in the AI community to prioritize data work — collecting, curating, analysing and otherwise considering the quality of data. But this is not practised as much as advocates would like, often because of a lack of institutional and cultural incentives. One way to encourage data work would be to reframe it as more technically rigorous, and thereby integrate it into more-valued lines of research such as model innovation.
For pride non-fiction reading may I suggest:
Ace by Angela Chen
and
Queer Data by Kevin Guyan
...they have shockingly similar covers...
When I get anxious about a research project that's not going very smoothly or quickly, I often try to remind myself that research is hard and if it all went super well then the problem probably wasn't challenging enough or wouldn't be considered research. But perhaps another framing is that research is a practice and I'm always learning and when something isn't going well it just means I'm learning a new thing.
Discovering that it's possible I like boring books. There's a particular kind of not-that-exciting book that I love. It's like abstract art or a movie that's all texture and no plot. But I don't like all boring books; I'm picky.
I like the word boring because it suggests bad and it's fun to turn the idea that boring is bad on its head. But of course boring might not be the right word.
slow calm gradual narrow inefficient sustained slight spacious steady undirected rhythmic even restless
Issue 2 of Crawlspace, a journal of digital-born literature, is out now! Best viewed on desktop:
https://crawlspace.coolOverall I had a shockingly good time at CHI, which in the past has felt super overwhelming. Something clicked this year, and that was really cool to experience.
- Finally, I met a lot of really cool people with similar interests. Technology for creative writing seems to be exploding, or at least getting a lot more popular than it was five years ago. But I don't really have the time to take on new collaborations, so it's a mixed bag. I'm happy to finally find my people! But it's a bummer to not be able to engage with them all very deeply right now.
- Work on creative writing reminded me of the tension between creative exploration of new technologies, which is rich and interesting, and the corporate exploitation many new technologies rely on. If I make something really cool and interesting and introspective with ChatGPT, am I effectively whitewashing the inherent problems of OpenAI? Maybe. This makes me sad, because I want to let creative exploration have free reign over technology.
- The difference between the paper session I presented in and my labmates paper session was palpable. I was in a smaller room with similar-ish papers; I felt I had a great audience and the room felt full and energetic. My labmate was in a large room with pretty random papers; he didn't get as many questions and the room felt empty. This was just due to the session we happened to be slotted into; I think our work was equally good!
rory green