The prompt never says Italy, but twelve layers into Qwen3.5-4B the model is already thinking it. J-Scope lets you watch that happen, inject France and get Paris, or drag one vector and watch the answer walk from euro to yen. A 72-second tour here; the live tool is one click away.

I wrote facts directly into a 4B model's J-space and KV cache on an RTX 4090. Three cache tokens matched full text-RAG, survived 48 distractors, and showed where retrieved facts actually live.

Can a cheap reward model tell a good AI agent plan from a bad one before you run it? I reproduced Orch-RM on a consumer RTX 4090 and RTX 3090: the verifier barely beats a coin flip on the signal that matters, yet still beats majority vote. Cheap to run, expensive to teach.