New paper with @PatrickButlin, from my time at @MATSprogram . We propose two new candidates for LLM individuation: the (virtual) instance-persona view and the model-persona view. 🧵 Are there multiple minds inside one LLM? Both views formalise the idea that the relevant individuals might exist at the persona level, which means a single LLM can host many minds.

New preprint ! We propose a new interpretation of DL relying on phenomenology in our paper “Rejecting Cognitivism: Computational Phenomenology for Deep Learning”, with G. Köstner and @ineshipolito. A 🧵 that conveys the key idea from an example. 1/11

arxiv.org/abs/2302.09071 We propose a novel formulation of computational phenomenology and apply it to deep learning. Doing so, we propose a new way of interpreting ANNs without relying on Turing-like representations. 2/11