LLM weights frozen; user facts not incorporated during inference
The post states that current LLM weights are completely frozen after training, meaning user-provided facts, corrections, or long documents are not integrated during conversation. This highlights a limitation in adapting to new information in real-time.