Summary
I’m exploring an idea for a new kind of platform: one focused entirely on publishing falsifiable conjectures and inviting refutations - structured around Karl Popper’s principle of knowledge evolving through bold claims and critical testing (conjectures and refutations). I’d love to know whether people in the EA ecosystem would find this valuable, or whether it’s doomed to be a high-effort intellectual graveyard. Open to hard critique.
💡 The Core Idea
A web platform where:
• Users post conjectures: clearly articulated, falsifiable ideas (e.g. “By 2030, bioengineered meat will be cheaper than conventional meat in high-income countries”).
• Each conjecture must include falsification criteria - what would prove it wrong.
• Other users post refutations: rational critiques, counterexamples, proposed empirical tests.
• Authors can revise their conjectures based on refutations - each revision is tracked.
• Long-lived, high-quality conjectures can be “published” (similar to journal elevation).
• Optional: prediction market layer (e.g. bet tokens on whether a conjecture will hold).
Think: Popper × LessWrong × Metaculus × open peer review.
🤔 Why I Think This Might Matter
• EA encourages rigorous thinking, public reasoning, and epistemic humility - but most platforms still favour persuasion over truth-seeking.
• The current discourse on forecasting (Metaculus, Manifold, etc.) is quantitative, but often lacks explanatory conjectures and discursive refutation.
• This could serve as a thinking tool and public knowledge archive for tracking evolving ideas on AI risk, biosecurity, global priorities, etc.
• It makes intellectual honesty a feature - not a liability.
🔍 What I Want to Know
• Would you personally use this? To propose ideas? Refute others? Browse?
• Is this meaningfully different from Metaculus + LessWrong + comments?
• What kinds of ideas would you want to test on such a platform?
• What design constraints would be critical to avoid it becoming low-signal or stagnant?
• What would make it emotionally rewarding enough to stick with?
🧠 Examples
Some example conjectures that could live on the platform:
• “Universal Basic Income reduces non-violent crime rates in high-income countries.”
• “By 2032, AI systems will autonomously generate at least one accepted pure maths theorem.”
• “Long-term caloric restriction slows human epigenetic ageing by ≥15% over 10 years.”
Each one would include:
• A short description
• Falsification criteria (specific metrics or evidence that would disprove it)
• Public discussion + potential revisions + token-based prediction layer (optional)
🧱 Status
I’m bootstrapping a simple MVP in Rails. Just trying to keep it lean. Still unsure whether to build it out or pivot, depending on interest and feedback.
🙏 Would Love Your Thoughts
This community understands the epistemic stakes better than most.
If this concept resonates - or sounds totally misguided - I’d appreciate any honest input, pushback, or reframing.
Would you use this? Why or why not?
X thread with screenshots: https://x.com/Duarteosrm/status/1909709276597149939
Thanks what's the core difference between this and forecasting?
Good question. The core difference is this:
Forecasting is about assigning probabilities to future events.
Falsification is about testing whether an idea can survive clearly defined attempts to prove it false.
Forecasting asks, how likely is this to happen?
Falsification asks, what would prove this wrong, and has that happened?
This matters because not every meaningful idea resolves cleanly into a forecastable event.
For example, “UBI reduces crime” or “MoND is a better fit than dark matter at low accelerations” are not yes-or-no outcomes with clean resolution dates. They are explanatory claims that require careful, falsifiable framing and rigorous testing - not just a probability score.
Scientists, institutions, startups, or EA orgs could publish hypotheses with explicit bounties for refutation. For example:
“We offer $500 to anyone who can provide a reproducible counterexample to this published claim under defined criteria.”
This flips the incentive structure. Instead of just publishing or forecasting, you’re paying to be proven wrong, and rewarding others for helping you find errors early.
For startups, this means posting falsifiable assumptions about product-market fit, growth loops, or user retention, and inviting outsiders to challenge them.
For EA orgs, it means exposing theories of change to public scrutiny, backed by incentives for constructive falsification.
It turns falsification into a public good, not just a peer review ritual. And it introduces a new tool for intellectual quality control: pay to test your beliefs.
Forecasting tells you what might happen.
Falsification tells you whether your thinking can survive contact with reality.
Both are valuable, but they answer different questions, and serve different parts of the truth-seeking stack.