GPT4: Eldritch abomination or intern? A discussion with OpenAI

Trust in Tech: an Integrity Institute Member Podcast

OpenAI, creators of ChatGPT, join the show! In November 2022, ChatGPT upended the tech (and larger) world with a chatbot that passes not only the Turing test, but the bar exam. In this episode, we talk with Dave Willner and Todor Markov, integrity professionals at OpenAI, about how they make large language models safer for all.

Dave Willner is the Head of Trust and Safety at OpenAI. He previously was Head of Community Policy at both Airbnb and Facebook, where he built the teams that wrote the community guidelines and oversaw the internal policies to enforce them.

Todor Markov is a deep learning researcher at OpenAI. He builds content moderation tools for ChatGPT and GPT4. He graduated from Stanford with a Master’s in Statistics and a Bachelor’s in Symbolic Systems.

Alice Hunsberger hosts the episode. She is the VP of Customer Experience at Grindr. She leads Customer support, insights and trust and safety. Previously, she worked at OKCupid as Director & Global Head of Customer Experience.

Sahar Massachi is a visiting host today. He is the co-founder and Executive Director of the Integrity Institute. A past fellow of the Berkman Klein Center, Sahar is currently an advisory committee member for the Louis D. Brandeis Legacy Fund for Social Justice, a StartingBloc fellow, and a Roddenbery Fellow.
They discuss what content moderation looks like for ChatGPT, why T&S stands for Tradeoffs and Sadness, and how integrity workers can help OpenAI.

They also chat about the red-teaming process for GPT4, overlaps between platform integrity and AI integrity, their favorite GPT jailbreaks and how moderating GPTs is basically like teaching an Eldritch Abomination.

Disclaimer: The views in this episode only represent the views of the people involved in the recording of the episode. They do not represent Meta’s or any other entity’s views.