Everything we have published, including null results and studies still collecting data.
We turn three internal features of GPT-2 small up through a fixed grid of doses and measure the dose-response, the side effects, how long a dose lasts, whether two doses add, and whether the model notices. The plan is preregistered and results are pending.