Featured AI2er: Pradeep Dasigi

Pradeep Dasigi is a Senior Research Scientist with AllenNLP

Pradeep Dasigi is a Senior Research Scientist with AllenNLP.

What put you on the path to your current role?

It all started when I attended a summer school on Natural Language Processing in my third (junior) year of college in India (15 years ago!). I had not heard of NLP before then and was fascinated by the idea of applying algorithms to process human languages. That led me to take many courses in NLP and Machine Learning during my Masters program at Columbia University. I later worked for a startup that provided Machine Translation services. I enjoyed doing that, wanted to learn more, and decided to get a Ph.D. I interned at AI2 one summer during my Ph.D. at Carnegie Mellon, and liked it here so much that I kept collaborating with folks here after my internship, returned for a second internship, and stayed full-time as a Research Scientist.

What’s the most surprising thing that has happened with your work at AI2 recently?

In my first couple of years at AI2, I worked on building benchmarks — datasets or tests that are used to evaluate specific capabilities of NLP systems. At that time, I thought of these mostly as steps towards something bigger, mainly because large language models have been getting better, surprisingly fast, and I imagined these benchmarks would get “solved” and be forgotten by the community soon. I am happy to see that some of my earlier work (particularly Qasper and DROP) is still being actively used for evaluating the largest of language models. GPT-4 has not solved them yet as of early this year. Let’s see what happens later this year!

What are you looking forward to with your work in the coming months?

I have been working on multiple projects around studying what kinds of things language models are reliably good at. These range from understanding how language models learn to perform new tasks to efficiently (i.e., without a lot of human effort) making them perform tasks we care about. I always look forward to the weekly project meetings with my collaborators where we gain little bits of understanding toward answering these larger questions. Some of this work will hopefully end up impacting the OLMo models and make it easier to align them with human preferences. I am very excited about that!

Pradeep with his wife, Modhurima, and daughter, Sanjana.

What is your favorite thing about working at AI2?

AI2 provides researchers the freedom to pursue their research agenda and also the resources to work on large-scale projects. This makes it a unique place to do AI research. I also appreciate that it is very easy for AI2 researchers to collaborate with academia. My collaborations with UW and other universities have been quite productive.

What do you consider the most underrated activity or place in Seattle?

I love that Seattle, particularly the areas surrounding AI2, are extremely bike-friendly and beautiful, making bike rides really enjoyable! One of my favorite routes is starting at Lake Union Park in South Lake Union and going up to Golden Gardens Park in Ballard. It is a relatively short ride (about 45 min) but you get great views of the lake and Puget Sound. As a bonus, you also pass a couple of nice breweries along the way!

The view from Golden Gardens Park.

What new hobbies or activities have you tried during the pandemic?

I always enjoyed reading fantasy, particularly those with detailed imagined worlds and cultures. During the pandemic, I got into conlanging — essentially imagining a culture and designing a language for it (for fun!). I also dabbled in calligraphy and learned some basics from watching videos online.

Listen to the NLP Highlights podcast and follow the team @ai2_allennlp.

Check out our current openings, follow @allen_ai on Twitter, and subscribe to the AI2 Newsletter to stay current on news and research coming out of AI2.