I recently co-founded Psyfy, which uses language models for psychology, and created the AutoGRAMS AI agent framework.
I have been researching language models since 2015. During my PhD at the University of Edinburgh, I developed multiplicative LSTM architecture, which Open AI used as their language model of choice in their early pre-GPT research on scaling language models . I also developed Dynamic Evaluation methods for adapting language models at inference, greatly improving the state of the art at language modeling at the time and motivating the use of gradient descent to store memories.
As a former Senior Research Scientist at Salesforce, I developed GeDi as a method to control generation from large language models that can, for instance, make text generations more friendly, and less toxic. I used my knowledge of controlling language model style to develop brand voice language model for Gucci that was shown to help their customer advisors improve their sales conversion rate. I also worked on the ProGen protein generation project to use language models to generate functional antibacterial proteins and help design antibodies targeting auto-immune disorders.