plebeian market didn't let me do the work i know how to do, that turned orly into a beast. i put too many things into it. things nobody asked for. with protocol nobody wanted.

i can just do this. and before lunch all done. since claude opus 4.6 the rate at which i can do this shit has dramatically accelerated. it doesn't catch everything, or fix everything, or not make mistakes. but you can just cycle that a few times and more gets found. the bigger context is critical. idk even why they restrict that so much anyway. it's just because the bigger it is, the more memory bandwidth is burned, the more power is burned, and they only care about a consistent rate of output. they don't want us directly teaching the machines.

anyway, on that subject, once i get my shit together, i've got something cooking that will blow your mind. a service will be added to smesh.mleku.dev that demonstrates it in action. plans are already made in exhaustive detail. an AI that directly learns from everything, uses tiny amounts of memory and bandwidth, and ... it laughs, it gets bored, it needs to nap for a few moments sometimes to optimise its memory.