Focus Fibre
I’ve been forking the ulisp-cardputer repository to include somethings I like. A gradual extension of the uLisp and some new C/C++ primitives are likely on the audio and GUI front.
I’m also thinking about AI cross attention. So optimizing the transformer attention as a “bundle” aggregate turns an I.M.O matrix problem into a I.A.M.B.O problem with M’s size reducing to the lower square, and A and B mapping from the larger M to a smaller one. Kind of making an attention fibre.
The dropout training on various M, depending on the focus context makes for a way of doing a sum of agents, by some selection or weighting of (maybe A) and B based on input I or output O, or even A.M.B (iterative short term memory).