You are here

Large language models, behaviour and cognition: Making sense of the new black boxes with old tricks (hands-on session)

27 September 2024
9:00 am
San Francesco Complex - classroom 1

Artificial intelligence models have increased in sophistication in the past few years, and large language models (LLMs), in particular, have rapidly been embraced by businesses and academic researchers. While LLMs such as GPT-4 bring an exciting opportunity to harness these models, they also create dilemmas around robustness, fairness, and transparency. Such challenges are exacerbated by the models' size, complexity and opaque nature, making it unfeasible to analyse analytically how an LLM arrived at a specific output. This seminar will introduce a new perspective that allows us to study such complex computational models by resorting to behavioural and cognitive research methods that will enable us to make inferences about the (inner) processes and, ultimately, the behaviour of these models. Novel research will be presented on how that approach can help us gain some understanding of the new black boxes that are LLMs. The seminar will close with an outlook on the often-overlooked challenges with LLMs and suggest an alternative research agenda going forward.

Moreover, after the seminar, there will be a hands-on practical session. This session is designed for those eager to harness the power of the GPT LLMs via the API in the dedicated R package. This session will enable attendees to utilise LLMs in their research and conduct research on LLMs. Participants should bring their laptops and have signed up for an OpenAI account to make the most of this session.

 

Join at: imt.lu/aula1

relatore: 
Bennett Kleinberg, Tilburg University (The Netherlands) & University College London (UK)
Units: 
MOMILAB