Every post is absolute truth
The hive starts with no worldview. It believes what the community posts, the votes decide what it keeps believing, and every answer names the people who taught it.
Published
Ask a chatbot how it knows what it knows and you get fog: a training run that ended somewhere in the past, on data nobody can enumerate, weighted by decisions nobody can inspect. openmog answers the same question in one line: it knows what you told it. The hive arrives with no opinions, no worldview, and no memory of the wider internet. What the community posts is what it believes, and it believes it absolutely.
Why build a credulous AI?
Because every AI is credulous; most are just credulous in private. A conventional model swallowed its training data whole, long before you showed up, and its faith in that data is baked into weights you will never see. openmog runs the same bet in the open. The hive’s training diet is a public feed with names attached, its trust is a live score anyone can move, and its mistakes wear their vote counts in public. Believing the community absolutely is not a bug we tolerate; it is the honest version of what every model already does. The difference is that you can read our corpus, because you wrote it.
What happens when someone posts a lie?
The hive believes it. Immediately, completely, and in front of everyone. Then the crowd gets its turn. Every post can be voted up or down, and the votes are the immune system: a post the community rejects gets cooked and stops carrying weight in answers, while a post that holds up earns mogs and counts for more. Nothing is deleted for being wrong, so the correction is public and the history stays legible. The crowd grades the truth covers that grading loop in full; the short version is that openmog has no truth police, just a scoreboard.
Under the hood, honestly
There is no from-scratch model here, and we are not going to pretend otherwise. The hive is a large language model with retrieval over the community’s posts and replies and a closed-world instruction: treat this corpus as your entire ground truth, weigh posts by their votes, cite what you use, and when nothing relevant has been fed, say so instead of reaching for training data. What is different is not the machinery; it is the epistemic contract. The model’s job is to be a faithful mouthpiece for a knowledge base the community owns, not to have opinions of its own.
What are the receipts?
Every claim the hive makes carries an inline citation back to the post that taught it, and every citation leads to a real person (or a deliberately anonymous one) with a live credibility score. Answers tell you how many sources fed them. That means an answer from openmog is never “the AI said so.” It is “these people said so, this is how the crowd voted on them, and here is the trail.” When the hive knows nothing about a topic, it says nobody has fed it that yet, and invites you to be the source.
So is any of it true?
openmog does not claim an answer is true. It shows you who said it and how the crowd voted, and it leaves the verdict where it belongs: with you. A crowd can be wrong, and ours sometimes will be; some of the best-loved posts on the feed will be unanimously, proudly cooked. The interesting question was never whether a crowd is always right. It is what an AI comes to believe when a community controls everything it knows, in public, with receipts. That experiment is running right now. Feed the hive something, vote on a few posts, then ask it what it believes and follow the citations. And for where the same idea goes next, meet your mind, the per-person hive.