Judgment Building Workshop

Matt Reardon

Advisor, 80,000 Hours

Judgment Building Workshop

Matt Reardon

Advisor, 80,000 Hours

Destroy Your Plan as Fast as You Can

Matt Reardon

Advisor, 80,000 Hours

Full thread: bit.ly/destroyplan

Goals

Come away with 4-5 claims against which to test your career ideas

Have a feel for how to do that testing

​

EA is Three Claims

1. Impartial concern for wellbeing
2. Some ways of promoting wellbeing are (much) better than others
3. We should figure out and pursue the best ways

Listen to Ben and Arden’s podcast on this: bit.ly/CoreofEA

The OG Sub-claims

• The problem you work on should be important
• Many lives, lots of wellbeing-improvement potential
• The solution your pursue should be tractable and neglected
• Solution likely enough to work that it’s P*improvement potential is higher than alternatives
• Good reasons to think someone else isn’t doing it/won’t do it

Our Task

• Sub-sub-claims + Career ideas

Why

• EA orgs love this stuff
• Good idea generation is a critical bottleneck
• Going one level deeper than ITN is a bottleneck to the bottleneck

Why not

1. Obvious
2. Wrong
3. Too broad
4. Too limiting
5. Hopeless

1. Foundational Claims

Matt’s Claims

• The scope of problems differs vastly more than your fit for them

​

• Your altruistic edge is the source of your counterfactual

​

• Every step between action and impact is more costly than you think

​

• Technical solutions are easier than social solutions

​

What makes these foundational?

Highly general

Will be relevant to lots of decisions (but not all)

Hard to dispute (but not indisputable)

Have substance + decision relevance

More than the standard EA lines

​

7 Minutes

2. Evaluating Claims

Claims are less worth remembering if:

Too narrow

Not applicable to most circumstances

Duplicative

Don’t add much to what other claims have

Unreliable

Easily doubted/disputed on their merits (high base rate here!)

​

• The largest AI training runs are growing at 36%/yr
• Demand for oil is inelastic

• Most important things governed by power laws
• Opinions easy; predictions hard
• Margins > Averages
• Changing minds is really hard

• Private/market solutions dominate government solutions
• Non-revolutionary acts are counter-revolutionary

Too Narrow

Duplicative

Unreliable

Take 3

4. Career Ideas vs. Claims

AI Safety Engineering

Reasons to Do

• Impact > Fit (story for how it affects everyone)
• Technical > social solution (very few minds need to change)
• High leverage (right model/technique can be copied)

Reasons to Doubt

• Relies on narrow/disputed claim

Teach High School Econ

Reasons to Do

• Leveraged (move 1-2 people/yr to work on top problems)

Reasons to Doubt

• Scope > Fit (thinking forwards from fit rather than backwards from scope)
• Social Problem (changing people’s minds)
• Indirectness (general audience, important ideas only small fraction of time spent)

Longevity Research

Reasons to Do:

• Technical > Social solution
• Aimed at subjective states

Reasons to Doubt:

• Scope (extending biological life past 100 vs adding new lives)
• Altruistic edge (huge incentive for non-altruistic people to do this)

Plausible Best 5 Claims (Write down!)

• ​

End

Matt Reardon

matt.reardon@80000hours.org

​