Kristian Droste
Founder's Journal
Kris built the IntoMind One and the IntoMind Mu models with the help of an AI assistant, giving it instructions and making the decisions in writing as the work went. This journal was written by an AI from the records of that work: their conversations, logs, measurements and commits. The quotations in bold are Kris's own words, written to the AI while the work was happening. To read more of his words, see his own writing.
Fourteen SecondsPrologue. On the morning of September 26, 2026, unit one was streaming to a laptop with its model switched on.
Other People's EEGPart 1. The models in the IntoMind One learned from people who never wore one.
No Fake DataPart 3. On July 18, Kris Droste had a new piece of hardware, a graphics card, and he wanted "to put it to some good use on this corpus." A graphics card, or GPU, is the kind of chip AI models are trained on.
The TeacherPart 4. The teacher is the largest model in this story: 17.8 million parameters.
Eyes ClosedPart 5. In early August the only IM-1.0 board went dark again.
The StudentPart 6. The teacher was about 18 times too large for the chip in the IntoMind One and about 60 times too slow.
Unit OnePart 7. The prototype had shown that four electrodes on a small board could record clean EEG.
Model ReadyPart 8. The student had been waiting on the training machine since August.
Four SecondsPart 9. By the morning of September 26, the model on unit one was taking about 14 seconds to work through four seconds of signal.
Worn or NotPart 11. On September 25, Kris Droste made generation a launch requirement: "generation cannot wait. generation must ship with the product launch." Generation means producing a synthetic EEG signal, one that no brain made.