A Platform for Falsifiable Conjectures and Public Refutation — Would This Be Useful?

post by PetrusNonius · 2025-04-08T21:09:23.819Z · LW · GW · 1 comments

Contents

  Why I think this might matter:
  Example Conjectures:
  Why it might fail:
  What I’d love feedback on:
None
1 comment

What would online discourse look like if it enforced falsifiability as a first-class constraint?

I'm considering building a simple platform. You can:

• Post a conjecture — a falsifiable claim, written clearly.

• Include falsification criteria — what observable data would disprove it.

• Others can post refutations — counterexamples, data, logical flaws, or proposed experiments.

• The author can revise the conjecture, with a full version history.

• Eventually: status shifts (e.g. “published”), prediction markets, and epistemic reputation scores.

The goal isn’t to predict things better than Metaculus or argue better than Twitter. It’s to treat falsifiability as a UX constraint and see what that changes.

Why I think this might matter:

Rationalist and forecasting spaces often emphasise Bayesian updating and prediction accuracy. But falsification, in the Popperian sense, is underrepresented. Refutation is rare. And most platforms reward agreement or persuasion, not structured disagreement.

This project is an experiment in building an environment where:

• The unit of discourse is a falsifiable idea.

• Intellectual honesty is rewarded.

• Revision is expected, not embarrassing.

 

Example Conjectures:

• “Universal Basic Income reduces property crime in high-income countries.

Falsification: Crime rates do not decrease in a 3-year UBI pilot, controlling for unemployment and inflation.”

• “By 2030, open-source LLMs will outperform closed models on math reasoning benchmarks.

Falsification: No open-source model outperforms GPT-N on GSM8K, MATH, or similar by EOY 2030.”

• “Long-term caloric restriction slows biological ageing in humans.

Falsification: No statistically significant difference in epigenetic age after 10 years of CR vs. control.”

 

Why it might fail:

• Nobody wants to be proven wrong in public.

• Writing falsifiable claims is hard, and refuting them is even harder.

• Intellectual work without emotional reward tends to stagnate.

• It could become LessWrong + prediction markets + versioning, but worse.

Still, I think it’s worth testing whether a norm-driven, falsification-first environment could push discourse toward clearer thinking and more epistemic accountability.

 

What I’d love feedback on:

• Would you personally post or refute ideas in this format?

• What would make this actually fun or rewarding to use?

• How do you distinguish this from Metaculus or LW posts with comments?

• What failure modes should I bake against from day one?

 

If there’s interest, I’ll share a working MVP soon. In the meantime, brutal critique is welcome.

X thread: https://x.com/Duarteosrm/status/1909709276597149939 

1 comments

Comments sorted by top scores.