๐ฌ As an AI researcher, I am enthusiastic about making LLM-based dialogue agents more human-like. To do so, I employ interpretability techniques and phenomena from psycholinguistics to uncover what makes language models tick in dialogue settings, then use these findings to propose potential vectors of improvement.
โ๏ธ As a research engineer in Cambridge, I craft and train custom LLMs for the legal industry. I am working on breathing life into an LLM-powered chatbot and "copilot" for lawyers that aims to provide a helping hand at every stage of a contract negotiation pipeline.
๐X/aronmolnr
๐ค HuggingFace/Arotte
๐ LinkedIn/molnaron
๐ฌ Google Scholar