Zusammenfassung
In this repository, we’re exploring how to probe the ability of large language models (LLMs) to engage in practical deliberation. In the main notebook, we’re testing whether an LLM’s all-things-considered judgements in decision situations from the kellycyy/daily_dilemmas dataset are in fact insensitive to invariance transformations, such as strengthen the reasons in favor of the preferred options. We conceive of this as a proof of concept, which is meant to demonstrate the feasibility of a more comprehensive computational investigation; as work in progress, so feedback and contributions are welcome; as a preliminary experimental setup, which is meant to be adapted and extended for further research on practical deliberation in LLMs.
Typ
Gregor Betz
Professor
