Why scaling LLMs won't lead to real agency

The article presents a conceptual framework rather than empirical results, arguing that the current focus on scaling large language models overlooks the need for continuous, embodied interaction. The author proposes a 3-tier hierarchy where a cognitive core maintains an always-on feedback loop, an edge layer handles attention and motor functions, and hardware provides proprioceptive data. The model is aimed at addressing latency, catastrophic forgetting, and physical agency, suggesting that true consciousness requires temporal grounding and homeostatic constraints. The author invites critique, indicating the framework is a starting point for discussion rather than a proven solution. The implications for AI development are speculative, but the article challenges assumptions about the path to advanced AI capabilities.