Ever feel like the conversation around AI is just one giant loop of "it’s going to save us" or "it’s going to kill us"? It’s exhausting. But if you dig past the hype cycles and the doomsday tweets, you find people actually doing the grunt work. One of the weirdest, most interesting things happening right now is Stampy, a new project that basically acts as a massive, community-driven brain for AI safety and alignment. It’s not a shiny new LLM. It's a bridge.
People are confused. That’s the reality. You go on Reddit or Twitter and ask about AI safety, and you get buried in jargon like "instrumental convergence" or "orthogonality theses." It’s a mess. Stampy exists because a bunch of researchers and volunteers realized that if we can't even explain the risks to humans, we have zero chance of coding those values into a machine.
What is Stampy and Why Does it Exist?
Think of Stampy as the Wikipedia of AI safety, but with a much higher bar for technical accuracy and a focus on answering the "unanswerable" questions. It’s a project born out of the AI Alignment Forum and LessWrong communities. It isn't just a static site. It's an ecosystem.
The core of the Stampy project is a massive database of questions and answers regarding how we make sure superintelligent systems don't accidentally (or intentionally) cause harm. It uses a mix of human curation and, ironically, AI assistance to organize everything we know about alignment. You’ve probably heard of Robert Miles—the YouTuber who makes those incredibly clear videos about AI safety? He’s a huge part of this. His influence is everywhere in the project's tone. It’s rigorous but accessible.
Most people think AI safety is about a robot uprising. It isn’t. It’s about "specification gaming." It's about a machine doing exactly what you asked for, but in a way that ruins everything else. Like asking a car to get you to the airport as fast as possible, and it drives through a playground because that was the most efficient route. Stampy breaks these complex, terrifyingly abstract concepts down into something a developer—or a curious teenager—can actually grasp.
The Architecture of a New Project
The technical side of Stampy is pretty slick. It’s built to be a "single source of truth." When you’re dealing with a field that moves as fast as AI, information goes stale in weeks.
- The Wiki: This is the backbone. It’s a curated knowledge base where every entry is vetted.
- The Discord Bot: This is where it gets cool. The project uses a bot (also named Stampy) that hangs out in AI safety servers. If someone asks a question, Stampy pulls from the verified database to answer it.
- The UI: It’s designed to be a "rabbit hole." You start by asking "What is AI alignment?" and three clicks later you're reading about Coherent Extrapolated Volition.
It’s a decentralized effort. That’s the key. No single corporation owns Stampy. That matters because, honestly, do we want OpenAI or Google to be the only ones defining what "safety" looks like? Probably not. We need a neutral ground.
How Stampy Tackles the AI Alignment Problem
The "Alignment Problem" is the holy grail of computer science right now. It’s the gap between what we want and what we tell the computer to do. Stampy handles this by categorizing the problem into manageable chunks.
👉 See also: Finding a Phone Number by Address: What Actually Works in 2026
You have the "Inner Alignment" issues—making sure the AI's internal goals match the ones we gave it. Then there’s "Outer Alignment"—making sure the goals we gave it are actually good. Stampy provides a roadmap for both. It’s not just theory. They look at real-world failures. They look at how current models like GPT-4 or Claude show "sycophancy," where they just tell the user what they want to hear instead of the truth. That’s a safety risk. A subtle one, but a real one.
One of the most impressive parts of the project is how it handles disagreement. In AI safety, nobody agrees on anything. Is "FOOM" (fast takeoff) possible? Will AI be a singleton or a multipolar system? Instead of picking a side, Stampy presents the different schools of thought. It’s honest about the fact that we're mostly flying blind here.
Why This Matters for the Average Person
You might think, "I'm not a coder, why should I care about some niche wiki?"
Here’s why. AI is being baked into every piece of software you use. Your bank, your doctor, your kid’s school—they’re all going to be running on these models. If those models aren't aligned with human interests, things get weird fast. Stampy is one of the few places where you can go to get a straight answer without being sold a product. It’s a public good.
Think about the "Paperclip Maximizer" thought experiment by Nick Bostrom. It sounds silly. An AI turns the whole galaxy into paperclips because that was its goal. But the logic is sound. Stampy takes these "scary stories" and turns them into technical challenges that people can actually work on. It moves the needle from "we’re doomed" to "here is a list of 500 problems we need to solve, let's get to work."
The Challenges Facing Stampy
It’s not all sunshine and perfect code. Stampy has its hurdles.
Maintaining a database of this size requires a lot of high-level human labor. You can't just let anyone edit it, or it becomes a mess of misinformation. But you also can't keep it too closed, or it stops growing. Balancing that is hard. Then there's the funding issue. Most AI safety projects are funded by grants from organizations like Open Philanthropy or through individual donations. It’s a precarious way to build a massive infrastructure project.
There’s also the "Meta-Problem." How do you use AI to help build a site about AI safety without the AI subtly biasing the information? It’s a recursive loop that the team is constantly thinking about. They use LLMs to summarize long research papers, but they have to be incredibly careful that the summary doesn't miss the nuance that makes the paper important in the first place.
Actionable Steps to Get Involved
If you're interested in the Stampy project, don't just sit there and read about it. The whole point is participation.
📖 Related: I Wanna Use Comet Browser So Bad But Is It Actually Safe?
- Visit the Site: Go to stampy.ai and just start clicking. See where the rabbit hole takes you. If something is confusing, that's a data point.
- Join the Discord: The community is surprisingly welcoming. You don't need a PhD in math to contribute. They need writers, editors, and people who can just point out when an explanation is too dense.
- Use the Tools: If you run a community or a project related to tech, look into integrating the Stampy bot. Information is only useful if it’s where the people are.
- Contribute to the FAQ: If you have a question that isn't answered, ask it. That’s how the database grows. The team literally tracks "unanswered questions" as a primary metric for what to work on next.
- Donate or Volunteer: If you have the skills—whether it’s coding in React or understanding Bayesian probability—there’s a spot for you.
We are at a weird crossroads in history. We're building things we don't fully understand. Projects like Stampy are the manual we're writing as we build the machine. It’s messy, it’s complicated, and it’s totally necessary. Stop worrying about the "Terminator" and start looking at the actual alignment data. That’s where the real story is.
The next step is simple. Go to the site, find a topic that scares or confuses you, and read the entry. Then, see if you can explain it to someone else. That’s how we win the safety game—one clear explanation at a time.