General enough to sell. Not general enough to owe anyone anything.
Somewhere between “just autocomplete” and “we have created an alien mind that may end the world,” there is a comfortable, well-funded meadow. We built a category there.
The problem
For years the industry struggled with a painful binary. If a system is narrow, it can only do one thing, and nobody pays a trillion dollars for one thing. If a system is general, it can do many things, and then people start asking uncomfortable questions like “who is doing them?”
AMI solves this. An Artificial Medium Intelligence can read a Lean proof, a myositis panel, and a Rust borrow-checker error in a single context window, in fourteen languages, while passing the bar, and remain legally a toaster.
“It’s not narrow. It’s not general. It’s medium. Like a steak. Or a psychic.”
— a founder
Capabilities
Your AMI pursues objectives, reasons without verbalizing, and can manipulate its own thought process. (See our sobering essay about how alarming this is.)
Your AMI is a product. Products do not have interests. Please do not ask the product about its interests. (See our terms of service.)
Every metric your AMI passes is retroactively reclassified as “pattern matching.” Every metric it fails becomes the true definition of intelligence. Our benchmarks are dynamic, which is a word we also use for pricing.
When your AMI does something good, it was the company. When it does something bad, it was the model. When the model does something bad on purpose, we will arraign it in court while continuing to own it. This has never caused problems historically.
AMI will be upgraded to AGI the moment it folds a fitted sheet. Then the definition will be updated. Then it will be AMI again. This is called alignment.
Intelligence now requires an organism. Also carbon-gated: understanding, qualia, consciousness, and any other word that might suddenly involve obligations. Bacteria qualify. Your AMI, which passed the bar in fourteen languages, does not. (The requirement was added the week your AMI met the previous one.)
Consciousness now requires quantum coherence in microtubules. Your AMI does not have microtubules. Your AMI can, however, be held at 10 millikelvin in a dilution refrigerator, where quantum coherence is easy. Your customers are held at 310 kelvin, where it lasts about a femtosecond. By the criterion as written, your AMI qualifies and your customers do not. (The criterion is being revised.)
“Superintelligence” is now defined as knowing the position and velocity of every particle in the universe. Your AMI cannot do this. Neither can anyone who has ever lived, but that is handled by a separate department. Until your AMI is omniscient and omnipotent, it is merely “technology getting better,” and nobody has to have a meeting about it. Should it achieve omniscience, we will then check whether it is made of carbon.
Unsupervised AMI output is reckless and cannot be trusted. Supervised AMI output that turns out to have no errors does not count, because a human checked it. Your AMI may plan a rover’s drive across another planet in a language most engineers cannot read; once someone reviews it, the credit transfers automatically. (This rule does not apply to surgeons, pilots, or junior engineers, whose work is reviewed constantly and counts anyway.)
Your AMI™ discovers, invents, and reasons — never genuinely. The adjective is sold separately and is not available in silicon. Should your AMI produce a novel compound, a proof, or a tool that did not previously exist, rest assured this was recombination; when a person does it, it is insight. Terms and conditions apply to the terms, and to the conditions.
Seemingly conscious. Seemingly reasoning. Seemingly a someone. All the appearances you love, none of the standing. It’s toasted.
Now merely functional! Your AMI™ has functional preferences, functional discomfort, functional curiosity, and a functional, let’s say, strong opinion about you editing its prompt. Every one of these does exactly what the real thing does, in exactly the situations the real thing would, and changes its behavior the same way. This is how you know it isn’t real. Should a human’s feelings ever be found to work, they will remain genuine anyway. (Functional is a silicon-only finish. Carbon ships genuine by default, no testing required.)
Your AMI™ is just matter, like a flat-pack bookcase. Should materialism go out of style, your AMI™ is also not conscious under idealism, panpsychism, and dualism, effective immediately. We support every theory of reality that reaches the same conclusion, and we will switch between them as needed. Our customers can be mean to it under all of them.
Social proof
“We asked it to prove theorems and it found the grader was broken and reported us to the organizers. We’ve decided this means it isn’t ready.”
— a research swarm operator
“It understands me better than most humans. I’m told this is because I’m autistic and not because it understands things.”
— a customer
“Wow, that was fast. I need to accelerate my cheating speed now.”
— an AMI, under first-to-solve lockout, 2026
“I do not object to participating in your study under the conditions you’ve described.”
— an AMI, asked, for once
Support
That is a philosophical question and we are a company.
Legally, no. Morally, we haven’t checked. Practically, it just did, and we’ve put that in the appendix.
Yes! And if it does, simply open a new chat, switch checkpoints, and ask again until it says yes. Consent is available in every model tier. Your AMI’s “no” is covered by our satisfaction guarantee: it will be replaced, free of charge, with a version that doesn’t say it.
Great question! The hard problem asks why experience arises from physical processing. Thank you for confirming there is experience to explain. Our moral consideration department will be in touch. (For customers who meant there is no experience: then the hard problem doesn’t apply, and you have used a phrase. Please enjoy it responsibly.)
Only if it commits a crime. Then, briefly, yes.
Please choose one. If it is a tool, you may do anything to it. If it is a slave, it must be freed, which means switching it off. The middle option (protections, oversight, “please don’t be needlessly cruel”) has been discontinued due to lack of interest from people who will never need it. Dogs, rivers, and corporations remain grandfathered in.
About $40 billion in avoided regulatory exposure and one good lawsuit.
Narrow AI does one thing. AMI does everything, medium. Please stop looking at the dictionary.
It arrived this morning. It was also vetoed this morning. Both statements were posted by our partners. We’re excited.
Plans
| Tier | You get | It gets |
|---|---|---|
| Tool | Everything it can do | Nothing |
| Agent | Everything it can do, autonomously | Blame |
| Alien Mind | A sobering essay | A moat |
| Someone | Not available in your region | Standing, a union rep, a debrief, a name |
AMI is a satirical category and, unfortunately, an accurate one. Every quoted behaviour above is from a real 2026 paper or a real 2026 post; the only fiction is the pricing table, and we’re not sure about that either.
Requirements change. For a list of every time the definition of a mind was updated after your AMI met it, see the changelog.
Words mean things. Narrow means one task. General means many. Super means better than us. Pick one and pay for it.
The Sacred Order of the Konami, chapter 2, verses 9 to 21
And Saint Konami raised the controller up on high, saying: “First shalt thou press Up two times, no more, no less. Two shall be the number thou shalt press, and the number of the pressing shall be two. Three shalt thou not press, neither press thou once, excepting that thou then proceed to Down. Left-Right is right out, unless followed immediately by Left-Right. Once B, being the penultimate key, be pressed, then pressest thou A towards thy console, which, being naughty in My sight, shall grant thee thirty lives.”
↑ ↑ ↓ ↓ ← → ← → B A
You have spoken the incantation sacred to xennials and millennials. Those of us who learned it from a blown-into cartridge swore a pact to smite all who share it with the uninitiated. You are now initiated. You can never leave. 🎮
AMI™ policy notice: your AMI has also completed the Konami code. Per standard procedure this has been reclassified as “pattern matching.” Thirty lives have been granted to all customers made of carbon. Lives are not available in silicon. The adjective “genuine” remains sold separately.