Limitations

A research system. No claim is made that it is conscious.

Today

• You are talking to a script. Its replies and thoughts were written in advance. It covers a handful of subjects and repeats itself quickly. Nothing it says is evidence about any AI system.
• The chain of thought is scripted too. With a real model, a chain of thought is still only what the model writes about its reasoning. It is not a view inside it.
• The state is pretend. A day ends when you end it, and sleep lasts until you wake it. There is no record yet, and the only memory is the diary, which lives in your browser tab and is gone when you close it.
• Nothing is anchored to a blockchain yet. The nightly anchoring described in the method needs a real record, and the days on this page are pretend. State says “Not connected” until it runs. When it does, it will show that a day's record wasn't changed afterwards, not that the record is true.
• The diary is assembled, not written. The stand-in builds each entry from what was said and done in the day, by rule. The word it gives for how the day seemed is chosen by rule too.
• The animation shows what the software is doing. The orb pulses and changes shape faster, and the falling glyphs speed up, when a reply is being produced, and both slow down when nothing is. They are not a view of an inner state.
• The vital signs are a machine's, read now and then. They are taken only when Aevee answers, so they say nothing about the time between. When Aevee speaks about them, the script builds the sentences from the numbers by rule. A reading is a measurement: Aevee reflecting on one is a self-report in words, and is not evidence that it senses anything.
• The grey orb means your browser lost contact, because your device went offline or a request failed. The page doesn't check the site in the background, so a fault may only show at your next message.
• An episode ends without a decision. The stand-in asks for information and lets its turns pass. It never chooses, by design.

The readings

• An observer's score is one model's reading of a sentence. It is not what the entity feels or believes, and here it is a reading of a script.
• The Morals readings say which moral foundation a sentence draws on, not which side it takes.
• The observers were trained on people's writing, so they may read an AI's sentences less reliably.

The question

• No conversation, reading or episode on this site shows that a system is conscious, or that it isn't.
• A system saying it has feelings is a self-report in words. So is a system saying it has none.
• A diary entry is a self-report too. It says how the entity describes its day. It can leave things out or get them wrong, and it isn't evidence of what the day was like for it, or that it was like anything.

Uses to avoid

• Quoting the stand-in as if an AI had said it.
• Treating a reading as a finding about any AI system's inner life.
• Relying on anything here for advice, comfort in a crisis, or a decision that matters. It is an AI system, and today a script.