Participate in amplification experiments
We're currently running weekly online experiments on amplification with humans, and we'd like to slowly expand the number of participants. If you're interested in participating, fill out this form.
- Our experiments are essentially about recursive question-answering in a setting where each user only observes small, isolated sub-questions, and is tasked with either answering them or breaking them down into further subquestions.
- We're starting with simple questions (estimation questions with access to Google, SAT-style reading comprehension questions) and will iterate until we get to a stage where a group of 10-20 people can reliably solve those questions. If that works, there are two follow-up directions: 1. Make the questions harder. 2. Try the same questions with proper blinding (i.e. a larger group).
- Logistically, we meet up in a Slack channel, then spend maybe 40 minutes question-answering (using Mosaic ), and maybe 20 minutes discussing how it went and how it can go better next time, with interleaved discussion and question-answering.
- Mosaic still has plenty of bugs and usability issues, so be prepared to suffer a bit. Between experiments, we will try to fix issues that come up.
- You're not committed to participating every week, and you can bail out at any time, but the experiments are most useful if there's a core group of people who stick around and are learning over time.
- We're in the early stages of learning how to run experiments, so things will probably be rough for a while. I'm very open to ideas for how to run experiments like this better (both logistics and experimental setup).
Are you usually free Thursday 5-6pm PST?
What other times would generally work well for you? (If Thursday 5-6pm PST works, feel free to leave this blank.)
How much time would you be willing to spend on experiments per week?
How many weeks do you think you'd participate in experiments?
More than 6 weeks
Which of the following have you read, if any?
Factored Cognition -
Iterated Distillation and Amplification -
AlphaGo Zero and capability amplification -
Humans Consulting HCH -
Strong HCH -
Supervising strong learners by amplifying weak experts -
Anything else you'd like us to know?
Never submit passwords through Google Forms.
This form was created inside of Ought.
Terms of Service