we do applied alignment research to accelerate epistemic capabilities and we are interested in the following research questions.
we are proposing to use AI to create epistemic tooling for the AI world. we would need to understand how the models relate to truth and our research agenda is mostly guided by the questions: can we build an external epistemic discipline around llms, such that their useful capabilities are harnessed while their unfaithfulness, overconfidence, sycophancy are understood and contained?