I have been busy interviewing AI safety experts about their "big picture" of the existential AI risk landscape: what's going to happen with AI, how might things go wrong, what should the AI safety community be doing, and how does their work fit in with that.

In the end I will have responses from around 20 experts, from DeepMind, Anthropic, Redwood, FAR AI, and others.

Some examples of questions I asked were:

• what will the first human-level AI look like?
• how could AI bring about an existential catastrophe?
• what are the most promising research directions in AI safety?
• what are the biggest mistakes the AI safety community has made?

The results will be published in the form of a post or sequence summarizing the results, probably transcripts of the conversations, and maybe audio recordings (would people be interested in transcripts and/or recordings?)

Pre-register your predictions of the most common responses to these questions here, here and here!

This work was funded by FAR AI.