My goal isn't to make LLM-assistance impossible; it will still be possible. In fact, GPT2-level inference is one of launch demos I have planned if I can finish this cursed self-hosting run.

My goal is to make training (especially self-training) impossible; while making inference deterministic by design and highly interpretable.

The idea is to build a sanctuary substrate where humans are the only beneficiaries of all possible technical advancements.