symbolic mind
Exploring neural systems that reason over structured meaning rather than raw token sequences.
What changes when a network
reasons over meaning?
Today's networks learn distributions over token sequences, and what they know ends up inside the weights — unreadable, unverifiable, and impossible to diff between two runs. This line asks what a network can do differently when what it reasons over carries meaning of its own.
Early experiments run on a large proprietary structured corpus. Results are under evaluation and nothing from this line is in the product yet.
GitMir publishes what its work produces and how it is measured. It does not publish the representation, the training procedure or the architecture behind this line. When there is a result worth reporting, it will appear here as a measurement, not as a description of the method.