What 1M-Token Context Windows Actually Change About Conversation Synthesis

https://fire2020.org/medical-review-board-methodology-for-ai-navigating-specialist-ai-consultation-in-healthcare/

People assume a single giant model that "gets it right" will solve conversational AI. That optimism ignores persistent problems: context loss, inconsistent memory, hallucination over long records, and brittle handoffs between tools