One of my favorite characters in Altered Carbon is Poe, the AI concierge who runs The Raven Hotel:
Poe is part of the hotel. He knows the place, understands its guests, answers questions, manages access—and, when necessary, can pull out the hotel’s machine guns.
I’ve built a less lethal version for this forums:
http://forum.objectivismonline.com
Their on-premises concierge is Gemma: Gemma 3 12B, running on my own GPU through LM Studio and integrated with Discourse AI.
Gemma knows what each forum is for. She has access to the forum’s content, mission, rules, and local context. She can search discussions, answer questions about the community, summarize and translate posts, categorize and tag topics, recommend related discussions, and pull out the guns to moderate new content.
Most websites put a CAPTCHA at the door and ask: Are you human?
I don’t particularly care.
My forums are explicitly built for humans and bots. Bots are welcome to crawl them, train on the public archives, answer questions, and even participate. The relevant question isn’t whether the participant is made of carbon. It’s whether the contribution belongs here.
A generic pet-bed advertisement on a forum about Ayn Rand’s philosophy is spam regardless of whether a human typed it. A thoughtful argument about Objectivist epistemology doesn’t become spam because an agent wrote it.
So instead of trying to exclude machines, I let a machine that understands the forum’s purpose help decide what belongs. Gemma inspects suspicious posts in context and has scoped authority inside Discourse: admit content, flag it for review, hide it.
Poe doesn’t run The Raven Hotel by asking guests to prove they’re human. He knows what hotel he’s running and acts accordingly.
Model + local knowledge + tools + permissions + institutional purpose = an AI operator.
Poe has The Raven Hotel. Gemma has a couple of philosophy & economics forums and a GPU in my house.
For now.
More on how the system is built: freecapitalists.org/how-its-built/