What we look for in proposals

There are some obvious criteria: good ideas, relevant to the topics of the conference, presented by qualified people. In addition, we are especially excited about:

Content that is best or only seen live. For sessions, this could be because it's new work, it's sensitive and the speaker wouldn't be able to share it publicly, it's a conversation where latent opinions or knowledge can be elicited, or because it's interactive. For demos, we're most excited about things like hardware, software that people can interact with, and art.
Sessions with hosts/speakers/facilitators who are good at leading sessions. For talks, this means good public speakers. For debates or fireside chats, this means a skilled moderator. For workshops, this often means the person has run the workshop before and has iterated on and improved it.

FAQ

How do content decisions relate to general admissions decisions?
When will I hear back about my proposal?
Do I need to submit an application to attend in addition to my session proposal?
If my proposal for a session where I am speaking is accepted, and I become a speaker for the conference, what is covered? Will I be compensated?
If my proposal for a demo is accepted, will I have to buy a ticket? How much do they cost?

Long list of questions

Here’s a long list of questions that we think it would be cool for content to address. This is meant to be purely inspirational, not restrictive; do submit proposals for sessions that don’t answer any particular question on here, but answer a question you think it would make sense to have on here!

📈 Capabilities

What capabilities would need to be unlocked for AI to meaningfully accelerate the process of AI R&D? How far off are those?
How fundamental is the gap between performance on benchmarks and real-world tasks? Is it growing or shrinking?
Will training for long-horizon or hard-to-verify tasks end up being a significant hurdle on the path to generalized systems? How feasible is real-world RL?
Will we get superintelligent "idiot savants" - AIs that dominate narrow tasks but can't learn new things efficiently?*
Is superintelligence possible? That is, on abilities like forecasting, creativity, and research taste, if the average human is a 5, and the best human is a 9, could an AI be a 1000, or does the scale cap out around 10?
Broadly: what’s up with robotics?
- Where’s the state of the art? (demos please!)
- What does the process of robotics R&D look like, and to what degree could a non-embodied AI researcher accelerate the process?
- Will the bottleneck to human-level robotics ultimately be software (perception, planning, decision-making under uncertainty) or hardware (dexterity, force feedback, speed)?
- How spiky are robotics capabilities? That is: how different are the timelines to robots that can reliably do e.g. warehouse work, agriculture, care tasks, etc?