Hi. I'm Yoshimi.
Welcome.
They are programmed to destroy us.
Defending against them — fighting them, protecting ourselves from them — is important work. Some people are doing it well. Many are not.
But we're focused on a different, less malicious, but more pervasive risk: undercooked robots.
We don't want to avoid it. We don't want to opt out. The fear of evil robots shouldn't push us into refusing what AI makes possible — including things humans alone could never make.
That's worth saying clearly. Most writing about AI is afraid of it or in love with it. Both stances miss the larger truth: this is one of the most extraordinary tools we've ever built, and we want to use it ambitiously, and the work of using it well is harder and broader than anyone is admitting.
The undercooked ones.
The robots that exhibit the behavior of evil robots without anyone meaning them to. The ones that ship with opt-out telemetry by default, or replace user tools without disclosure, or get trained without diverse voices in the room, or optimize for one metric while quietly breaking three others.
Their builders aren't villains. The builders are usually thoughtful people working in systems that didn't require enough cooking before shipping.
The robots are still dangerous.
When the pink kind are normal — when undercooked is the default — the red kind become harder to spot, easier to build, and more socially acceptable.
The work of cooking your pink robots more thoroughly is upstream of the work of stopping the red ones. Every well-cooked robot raises the floor for everyone.
That's why this site exists.
One. AI reflects us. The work of shaping what it becomes can't be delegated to a small group of experts. It's collective work, and most of the people who should be doing it haven't been invited yet.
Two. Evaluating AI tools shouldn't require being an expert. Anyone using AI — to build with, to buy, to depend on — needs usable ways to ask good questions. We make those.
Three. AI shaping needs more than computer scientists. The kindergarten teachers, therapists, ethicists, lawyers, artists, doctors, and translators of the world have decades of intuition about how to shape behavior in other minds. That's exactly what training AI is. The hiring pipelines at frontier labs don't reflect this yet. They should.
Four. Use AI to extend what humans can create, not to replace what humans should do. The most valuable AI use is the kind that lets you make something that couldn't exist without it. Not faster output of the same thing — new categories of thing.
This site is new. We're starting small.
— A prompt you can paste into your AI of choice to evaluate any AI tool you're considering or building. It asks the questions a careful person would ask. It runs on your machine. It doesn't talk to us. (Coming soon.)
— Notes on what well-cooked AI work looks like in practice. Including some of our own. (Coming soon.)
— A working belief that the principles for what cooked AI means should be developed openly, by many people, over time. We're early in figuring out what that looks like.
We don't certify anyone. We don't issue badges. We don't tell you which tools to install or avoid. We give you better questions to ask and better frames to think with, and we trust you to use them.
We're also not trying to slow AI down or speed it up. We want it to be cooked.
Good Robots is an early project. Right now it's mostly one person, with help from a small group of people who care about this. It's named after the Flaming Lips song “Yoshimi Battles the Pink Robots,” which we've been listening to for years and only recently started hearing differently.
The work of cooking robots well needs more disciplines than computer science.
If you bring expertise from somewhere else — teaching, therapy, medicine, law, design, religion, addiction recovery, conflict mediation, social work, parenting, journalism, art, anything — and you want to develop and own the perspective from your field on what well-cooked AI would mean, reach out.
Tell us:
— What discipline you bring.
— What you think your field knows that AI builders are missing.
— What you'd want to contribute or own.
We don't have a formal program yet. We have a project, a mission, and a small group figuring out what comes next. The right people in early will shape what that becomes.
— Yoshimi