Learnings from Building Epistemic Tools

My work at Goodheart Labs

Rob¹ and I have built a number of tools at Goodheart Labs. Here are some learnings, written down in about an hour².

People don’t like extra thinking

Epistemic tools already exist. You have been able to google anecdotes for years, but most people don’t. Now it’s even easier. I can use Perplexity to check things with a higher success rate. I doubt most people still will.

If people are writing a tweet or telling an anecdote, I reckon the “checking” process has to be sooooo cheap for them to do it. The underline feature on word processors is good. I guess twitter’s “are you sure you want to send that sweary tweet” works. But in general I don’t think people can be bothered to uses even quite simple processes.

I write more about that here:

Careless thinking

Many ideas are good in theory

We’ve built some neat tools. The visualisations we built for findingconsensus.ai. The voting process for viewpoints.xyz. A whole forecast question generation site³.

Initially there was often interest, but these ideas didn’t actually get traction when built.

I suggest prototyping quickly (and getting funding that supports this).

Dashboards work, but have their moment

We built BirdFluRisk.com and it got a solid amount of traffic. I think had bird flu become worse (or nearly dones so) it would have gotten even more.

Dashboards seem to me to work if built at exactly the right moment when many people are seeking to make sense of something.

I wouldn’t be surprised if this is true of many epistemic tools (that are often offered for free, rather than a subscription)

There are many successful epistemic tools already

• Google search

• Spreadsheets

• Wikipedia (but you probably know that)

• LLMs in chat interfaces

• Anki

• Infographics

• Prediction markets

• Community notes

• Notion, Roam, Obsidian, Logseq, etc

• Reference manager

• Reddit, Quora

I think it’s sometimes easy to forget things that already exist. These days I’m more interested in widening tools that already work than trying to build entirely new ones.

Can you do it with a google doc/ airtable/ prompt/ vibecoded-app?

Many things can be quickly prototyped in existing tools before needing to get down and dirty building proper infrastructure.

We built Doubtful, a beautiful prediction market question generation app, but noone ended up using it. I spent wayyy too much on developing that. I should have tried getting a group on an airtable first.

This is an obvious learning but I wish I always thought it. Actually I’m gonna go make a TAP for it.

What do you think?

What do you think of this? Do you have learnings of your own? Please send them in the comments.

1. My lead developer.

2. I’m going to a conference and I realised this would probably be valauble

3. Doubtful is no longer live