OpenAI5Google DeepMind7Anthropic1
Geoffrey Irving· from 2:56 in the full interview
This website is a project of Palisade Research.
Palisade Research is a nonprofit whose mission is to help people and institutions build the understanding needed to avoid permanent disempowerment by strategic AI agents. To learn more about this project, see the FAQ.
frominside.ai is a collection of interviews with current or former employees of frontier AI companies, in which those employees express their personal views about AI development.
The massive response to Jacob Coxon’s and Evan Hubinger’s tweets showed that there is a lot of demand from the world to hear from employees at frontier AI companies about what is happening with the technology.
We believe the general public does not realize that the people building frontier AI earnestly consider, or even take for granted, truly extreme possibilities like some of those shared in this series.
Our goal with this project is to open up a direct channel from the people working on this technology to the public.
We at Palisade generally agree with Jacob Coxon that the AI companies are “gambling with our lives.” We think that few people in the world understand the situation we’re all in, but if the public did understand, human civilization would have a better chance of reaching an amazing future.
We were only able to film interviews with our subjects in London because Andrew Draganov heard about our project and offered to repurpose a film shoot that he had scheduled for a different project. The London team managed all of the logistics and filming, and Eli conducted the interviews via Zoom.
The frominside.ai website was produced by Lightcone Infrastructure.
We especially thank every interview subject who filmed with us. This project would obviously not have been possible without your participation, and we salute you for speaking up about what you believe and why.
Rosie Campbell, Juan Felipe Cerón Uribe, Geoffrey Irving, Andreas Kirsch, Daniel Kokotajlo, Victoria Krakovna, Jeffrey Ladish, Vishal Maini, Neel Nanda, Mary Phuong, Jeremy Schlatter, Alex Turner
We contacted people through our networks inside the AI companies, colleagues they recommended us to, and employees who had spoken publicly after Jacob Coxon’s resignation. We generally sent them a message similar to this invitation, and encouraged them through personal correspondence.
Our interview subjects are not a representative sample of AI company employees and former employees, for at least two reasons. First, we reached out to people through our networks, which introduced some selection effects. Second, the people who agreed to be interviewed were more likely to be those who felt they had something important to share about the AI situation. In practice, they were disproportionately employees on safety teams, people who are more concerned about the risk, and people who are more enthusiastic about pacing the frontier. For a more representative sense of what AI researchers think, see AI Impacts’ most recent survey of AI researchers.
We filmed 22 interviews; many will be released only if the person later gives permission.
Interviews were unscripted. Subjects received a list of default questions beforehand, which told them they could decline any question, could do multiple takes, and we edited for flow and clarity, sometimes joining takes.
[This is a public doc. Feel free to share with anyone.]
Note that these aren’t necessarily the only questions we’ll ask. In particular, we’ll ask follow-up questions to your answers—basically trying to ask the questions that we imagine a member of the public would have if they were listening to you. But these basically cover the topics that we want to ask you about.
You’re also free to decline to answer any question. Further, if you think that there’s something else that’s important and not captured by these questions, we’re glad to film that instead or in addition. These are intended as scaffolding, not requirements.
Some sample questions to ask lab employees and former lab employees
The invitation planned filming in a rented studio in downtown San Francisco; the London interviews were filmed by a local crew, with Eli interviewing over Zoom.
If you want a more representative sense of what AI researchers think about the future of AI (or at least what they thought in late 2024), you might be interested in AI Impacts’ most recent survey of AI researchers publishing in top venues.
All the videos are released under the Creative Commons Attribution 4.0 license. Clip away!
Two reasons: First, our desire to have participants’ authentic perspectives means that we tried to have a light hand as we edited raw footage.
Second, because we had never done this before, and though we worked with some incredible producers and editors, there were only 15 days between when this project was conceived and when we launched.
One thing any American can do: call your congressperson and tell them about your concerns. Members of Congress are very sensitive to what their constituents care about, and they gauge that partly through receiving calls from voters. You can use callcongress.ai to get help on how to call effectively, including preparing a script for your call.
(If you live in any other country, we encourage you to express your voice there.)
If you want to go deeper, you might be interested in this post from 80,000 Hours titled “How to get into AI safety in 3 months”.
As Palisade’s Executive Director said in his video, “The future is not yet decided. We have choices to make.”
We are still filming more interviews! If you work for, or have worked for, a leading AI company, and you want to contribute an interview or a (non-interview) statement to this project, you can get in contact with us here.
We would especially love to hear from capabilities researchers or employees who are skeptical about the risks, so that those perspectives are also represented in this project.