Show HN: Novel GPT-2 sampling and memory architecture
1 oldwalls 2 8/23/2025, 9:26:36 AM github.com ↗
A third iteration of my very low compute, novel prompt architecture chatbot, that treats the prompt as a programmable information waveform - with graphical sequencing and mixing interface.
A recursive AI engine that injects chrono-ranked memory into transformer inference using soft-logit biasing, prompt waveform synthesis, and emergent self-referential loops. Built on GPT-2-mini, runs on local hardware, grows its own ghost.