Pontus Stenetorp: LLMs and Future of NLP

Natural Language Processing (NLP) as a field (as well as Artificial Intelligence in general) is currently seemingly undergoing a phase of radical change. The reasons for these changes are several, but key factors are: recent breakthroughs in Large Language Model (LLM) performance, the rapid commoditisation of NLP systems as direct solutions to business problems, and ever-increasing computational requirements to “keep up” with the state of the art.

In this rather unorthodox talk, I will briefly touch upon recent research accomplishments of myself and my group in the context of LLMs in terms of performance and behaviour, while relating these to predictions for the models that have appeared over the last twelve months. Following this, I will move to a more speculative portion of the talk, where I ask and attempt to answer questions such as: What research questions and methodologies are likely to prove fruitful in the near future? What should (and could) our role as NLP researchers be relative to society, other fields, and industry? How do we best organise ourselves as research units and institutions given the changes that we are experiencing? My hypothesis is that we may need to fundamentally reconsider which tasks we consider relevant and how we approach them. That we better need to take into account serving societal and research (as well as practical) needs outside our immediate community. All while we likely must form larger consortia and reconsider how we organise ourselves within our research groups. I will also give concrete examples of actions taken along these lines within the UCL NLP group.

I should emphasise that I, of course, lack definite answers to these questions. However, based on my own experience leading a world-leading NLP research group I have found it impossible to effectively lead the group over the last few months without at least attempting to answer these kinds of questions to be able to provide a “vision” for my fellow academics, researchers, and students. It is thus my hope that by sharing these thoughts with a wider audience that others can benefit from my own perspectives and – hopefully – provide me with perspectives of their own.

Pontus SAITO STENETORP is an Associate Professor in Natural Language Processing (NLP) at the University College London (UCL) Department of Computer Science, Deputy Director of the UCL Centre for Artificial Intelligence, and UCL-Facebook AI Research PhD Programme Director. They lead the UCL NLP group and their primary research contributions and interests lie in the areas of NLP model and system development, evaluation, and analysis. Their research has received awards at leading conferences in their field: outstanding paper at EACL 2017, best paper at EACL 2021, and outstanding paper at ACL 2022. To date they have published more than 80 publications, which have been cited more than 4,800 times.

Previously they have been a Postdoctoral Research Fellow at the University of Tokyo and Special Project Research Associate at the National Institute of Informatics in Tokyo, Japan. They received their PhD in Computer Science from the University of Tokyo in 2013 and MSc in Computer Science from the Royal Institute of Technology (KTH) in Stockholm, Sweden in 2010.