Researchers proposed the Infinite-Parameter LLM architecture. A compact hypernetwork turns live data into a low-rank modulation of a shared base network. Weights are generated from live data rather than stored in a fixed bank. The model carries a Bayesian belief over the generator’s latent code and updates it online. Effective weights are re-derived from that evolving belief as the session proceeds. Stored footprint stays fixed while the weights the model can compile are effectively infinite. Knowledge and behavior supplied at run time are carried in the weights rather than the prompt. This is amortized in compute, frees the context window, persists across turns, and can generalize better than in-context use. The paper specifies an evaluation protocol testing this against in-context learning and retrieval.