Some rough notes from Michael Aird’s workshop on project selection in AI safety.
Tl;dr how to do better projects?
Backchain to identify projects.
Get early feedback, iterate quickly
Find a niche
On backchaining projects from theories of change
Identify a “variable of interest” (e.g., the likelihood that big labs detect scheming).
Explain how this variable connects to end goals (e.g. AI safety).
Assess how projects affect this variable
Red-team these. Ask people to red team these.
On seeking feedback, iteration.
Be nimble. Empirical. Iterate. 80⁄20 things
Ask explicitly for negative feedback. People often hesitate to criticise, so make it socially acceptable to do so
Get high-quality feedback. Ask “the best person who still has time for you”.
On testing fit
Forward-chain from your skills, available opportunities, career goals.
“Speedrun” projects. Write papers with hypothetical data and decide whether they’d be interesting. If not then move on to something else.
Don’t settle for “pretty good”. Try to find something that feels “amazing” to do, e.g. because you’re growing a lot / making a lot of progress.
Other points
On developing a career
“T-shaped” model of skills; very deep in one thing and have workable knowledge of other things
Aim for depth first. Become “world-class” at something. This ensures you get the best possible feedback at your niche and gives you a value proposition within larger organization. After that, you can broaden your scope.
Product-oriented vs field-building research
Some research is ‘product oriented’, i.e. the output is intended to be used directly by somebody else
Other research is ‘field building’, e.g. giving a proof of concept, or demonstrating the importance of something. You (and your skills / knowledge) are the product.
A specific process to quickly update towards doing better research.
Write “Career & goals 2-pager”
Solicit ideas from mentor, experts, decision-makers (esp important!)
Spend ~1h learning about each plausible idea (very brief). Think about impact, tractability, alignment with career goals, personal fit, theory of change
“Speedrun” the best idea (10-15h). (Consider using dummy data! What would the result look like?)
Get feedback on that, reflect, iterate.
Repeat steps 1-5 as necessary. If done well, steps 1-5 only take a few days! Either keep going (if you feel good), or switch to different topic (if you don’t).
Some rough notes from Michael Aird’s workshop on project selection in AI safety.
Tl;dr how to do better projects?
Backchain to identify projects.
Get early feedback, iterate quickly
Find a niche
On backchaining projects from theories of change
Identify a “variable of interest” (e.g., the likelihood that big labs detect scheming).
Explain how this variable connects to end goals (e.g. AI safety).
Assess how projects affect this variable
Red-team these. Ask people to red team these.
On seeking feedback, iteration.
Be nimble. Empirical. Iterate. 80⁄20 things
Ask explicitly for negative feedback. People often hesitate to criticise, so make it socially acceptable to do so
Get high-quality feedback. Ask “the best person who still has time for you”.
On testing fit
Forward-chain from your skills, available opportunities, career goals.
“Speedrun” projects. Write papers with hypothetical data and decide whether they’d be interesting. If not then move on to something else.
Don’t settle for “pretty good”. Try to find something that feels “amazing” to do, e.g. because you’re growing a lot / making a lot of progress.
Other points
On developing a career
“T-shaped” model of skills; very deep in one thing and have workable knowledge of other things
Aim for depth first. Become “world-class” at something. This ensures you get the best possible feedback at your niche and gives you a value proposition within larger organization. After that, you can broaden your scope.
Product-oriented vs field-building research
Some research is ‘product oriented’, i.e. the output is intended to be used directly by somebody else
Other research is ‘field building’, e.g. giving a proof of concept, or demonstrating the importance of something. You (and your skills / knowledge) are the product.
A specific process to quickly update towards doing better research.
Write “Career & goals 2-pager”
Solicit ideas from mentor, experts, decision-makers (esp important!)
Spend ~1h learning about each plausible idea (very brief). Think about impact, tractability, alignment with career goals, personal fit, theory of change
“Speedrun” the best idea (10-15h). (Consider using dummy data! What would the result look like?)
Get feedback on that, reflect, iterate.
Repeat steps 1-5 as necessary. If done well, steps 1-5 only take a few days! Either keep going (if you feel good), or switch to different topic (if you don’t).
Writing hypothetical paper abstracts has been a good quick way for me to figure out if things would be interesting.