When to compete (t-shirt design, ai art gen)? When to hire & grow?
- Better when evaluation is cheap
- Successful prize competitions are rare (?)
- Examples
- Hiring as a competition in a hiring round
- Hackathons -- many projects fail, but a few make it up (?)
- Object-level competitions can be good for hiring eng, traders, talent
- eg Marcus, Joel on Manifold; David on Manifest
Babble & prune for yourself
- Good to list out project ideas, or solutions for a problem, weigh out, and then pick. Breadth-first search
Select vs farm
- When to select? When you can only have one (eg partners for monog; open roles due to context & money; t-shirts when unified for a conf)
- When to farm? Manifund, Manifold -- websites can have many different contents, cater to a long tail, encourage plurality
- => against the EA implication of "a single best". Which is already being loosened by cause area considerations
- (select vs farm is spectrum, ofc -- Mox does some selecting and some farming)
generation vs selection
- what domains is generation cheap? AI research & writing & art; names;
- Does cheap generation imply expensive selection? Selection costs are higher, anyways to filter through everything.
- but what about on a per-item selection/evaluation cost?
- How do LLMs internally manage selection vs generation?
- Scaffolding paths often do a "try 3 and pick 1". (does this happen in chain of thought)
- Generating for a tournament/competition (like a hiring round) involves marketing -- often scarcest resource
- that's what bounties are trying to solve, though "how much to put on bounty" is hard. More $ doesn't scale to better value
ideas are cheap to generate connections, referrals, contacts are (sometimes) cheap to generate (Taren seems good at these)
Manifund examples
- Example: grants
- Manifund generates 20 projects a month?
- Evaluating is actually pretty fast (for Austin)
- But maybe a newsletter that surfaces 5 good ones every month would be pretty good. (like EAF digest)
- Example: regrantors
- Example: donors
- (probably: should user interview donors more)
- (probably: ask Jesse to start making grants)