Next-gen “community”, unifying Mox/Manifund/Manifold/Manifest:
Motivation:
- Breaking out of the “Dunbar’s number” limit on building community & running events
- Extend the snap judgement thing one does when scrolling through Partiful and deciding who to invite
- Many good event organizers want to find relevant people (cf METR hackathon, Ricki’s events)
- Imperfect modeling of others leads to problems with sales, marketing (cf passing along leads for events)
- Manifold & Manifund create salience, interactions, knowledge of people
Pitch:
- Build an extremely detailed view of who you are and what you want, based on your public interactions, writing
- As the Christian God: all-knowing, all-caring, unique relationship with you
- (When they say “building god” many mean “supersmart”, but God is more than that)
- Be able to hold all of this within a coherent system, query-able, powered by LLM
- High-fidelity simulations of your persona?
- Preserves your attention when you have too many opportunities to take
- See: Workshop Labs or Worth the Candle utopia assistants
- CRM on steroids; highly targeted opportunities
- Consider the data Google/FB/etc has on you. But: Google only monetizes this to the tune of $100/pp/y via ad spends.
- Deep knowledge of people should be worth so much more than that.
- E.g. Manifold Jobs; high value in knowing rates of growth
- Rather than very specific opportunities fragmented across different verticals (jobs vs shopping vs information vs love), could be unified, building a more holistic picture of people
- (maybe this is the direction OpenAI is going down?)
- names
- “meta-social-network”
- online presence
- whoami
- lodex — lodex.ai is available
- good: rolodex, lodash, lode
- bad: a bit 1990s
- uniq
- query
Imagine:
- The superpowered matchmaker that can immediately find your partner for life
- Recommends events like Manifest or communities like Mox for you
- Finds you cofounders, jobs, roles, next steps
Potential MVPs:
- Mapping out the EA ecosystem (individuals, orgs) based on Manifund/Mox/Austin’s rolodex
- Lightest version is a tool that’s just useful for me
- Building out a thing for ACX Grants, helping w/ review
- Consolidate the 50 CRMs we have right now
- Helping with our own pipelines for eg Manifold hiring, Mox fundraising, Mox Populi, etc
- Imagine auto-populating a CRM from existing data in one go. Airtable-like. Query the chat “Find me recommendations for people who can do Manifold eng”
- “Suggest people I know who I should invite to Manifund Regrantors Showcase”
Questions:
- Public data, vs private data?
- One useful thing is the web of links (Linkedin, socials, blogs), all scraped and ready to consume. Like Wikipedia for individuals.
- Another useful thing is the private info that only certain people have
Extensions:
- Mapping personas for orgs and communities too
- Could be meme-y: “Mox-chan wants more events”
Appendix/Musings:
- Event series to host: boring business skills
- Build a CRM; send cold emails; structure a standup
- (in the age of LLM?)
- Should always be automating your job.
- How to create more Manifolds? Create more Moxs?
- Or: Create more The Curves?
- (or: improve efficiency of existing systems)
- Bottleneck is sometimes more reachouts, accountability
- There’s value in doing things by hand — mostly, in that the things you supervise/own are the things that you can more easily modify, have loaded into your context
- “What to load into your context”
Convo with Claude, my takes:
Let's go with:
- Public data only to start: Linkedin, socials, Manifold/Manifund public profiles
- Start with the extended EA/AI safety scene. (So probably EA Forum/LW are high value too, as well as personal blogs)
- Our existing CRMs are almost all Airtables.
- I'd want this thing to stay automatically up to date
I'm wondering what a generalized approach is to storing this kind of data. I imagine it'd start with records for each individual, a set of associated links, and then we'd scrape that, store it, and provide eg a daily snapshot?
I also want searching over these to both be LLM-accessible (so: vector?) and also fast. Maybe using super fast inference things like Cerebras to query and run results?