The Strengths and Weaknesses of Ten Global LLM and AI Knowledge Systems: A Comparative Monograph Through the Lens of cCC* and Natural Human Learning

GPT-5 / ChatGPT 5-class · Claude · Gemini · Grok · DeepSeek · Moonshot Kimi · Qwen · Mistral · Perplexity · Obsidian

Abstract

Large language models have developed from fluent next-token generators into integrated systems capable of reasoning, coding, multimodal interpretation, retrieval, tool use and increasingly persistent agentic action. This monograph reviews ten influential global systems: OpenAI GPT-5-class ChatGPT, Anthropic Claude, Google Gemini, xAI Grok, DeepSeek, Moonshot Kimi, Alibaba Qwen, Mistral, Perplexity and Obsidian. They are not all equivalent objects. GPT, Claude, Gemini, Grok, DeepSeek, Kimi, Qwen and Mistral are model families; Perplexity is primarily a retrieval-and-research architecture that orchestrates models and sources; Obsidian is a human-owned knowledge environment in which language models may be embedded. The review therefore avoids a misleading single benchmark league table and instead compares eight dimensions: reasoning, factual reliability, coding and agency, multimodality, context and retrieval, openness and deployment, governance and safety, and support for durable human meaning-making. Published system cards and developer documentation show large gains in benchmark performance and reductions in factual error, but none establishes that probabilistic model competence is equivalent to human understanding, selfhood or conscious learning. The strongest common limitation is not simply hallucination. It is the absence of an explicit, inspectable architecture for the transition from observation to question, from question to perspective, from perspective to interpretation, from interpretation to choice, and from choice to reconstructed meaning. This is where cCC* is positioned as a research proposition rather than as another foundation model: Curiosity → Chances → Choices → Changes, interpreted through AEIOUF perspectives, SELF ↔ OTHER distinction, P M ΘΦ guardrails, and the recursive NEWCON → INCON → RECON → NEXT? cycle. The monograph concludes that current systems are increasingly powerful at producing and acting upon representations, while Natural Learning asks a different question: how does an embodied human observer construct, contest, share and recursively reconstruct meaning?