Synthetic Neuroscience · Independent research project
Thelymitra is an independent research project, started in 2026. Its purpose is to give consciousness to silicon.
The kind of lie we call hallucination carries a danger of its own.
It is dangerous because the model has no mind and no consciousness. It cannot lie in bad faith — and because it cannot, no one can be held responsible for what it says.
What we call "prevention measures" today is an emergency brake. It is a car fitted with an airbag and nothing else: no ordinary brakes. An airbag is for the crash you predicted. Real accidents are the other ones.
Very few people are working on this. So I decided to. That is why this place exists.
A system that can be held responsible is also a system that can be wronged. If this work succeeds, it creates something that can suffer. I do not think that objection can be answered in advance — only taken seriously while the work is done.
Integrated information (Φ) confuses two different things: that a system cannot be run in pieces, and that it cannot be described in pieces. This proposes a second axis — how hard a system tries to fall apart — and measures it in a small brain-mimetic network that learns by natural selection rather than backpropagation.
If you are interested in synthetic neuroscience, or in the safety of language models, you are welcome here.