LLM on C.CUI's Log

LLM on C.CUI's Loghttps://cuicaihao.github.io/tags/llm/Recent content in LLM on C.CUI's LogHugoen-AUThu, 04 Jun 2026 07:00:00 +1000No More 'Playing It By Ear': Why Harness Engineering is Becoming Key to AI Production Deployment?https://cuicaihao.github.io/posts/2026-06-04-no-more-playing-it-by-ear-why-harness-engineering-is-becoming-key-to-ai-production-deployment/Thu, 04 Jun 2026 07:00:00 +1000https://cuicaihao.github.io/posts/2026-06-04-no-more-playing-it-by-ear-why-harness-engineering-is-becoming-key-to-ai-production-deployment/Harness Engineering is emerging as a crucial paradigm for deploying AI into production environments, moving beyond unreliable prompt-based approaches. It conceptualizes large models as powerful “wild horses” and introduces a “rein system” comprising constraints, guidance, verification, and correction. This system enables AI Agents to accomplish tasks more stably, controllably, and reliably in complex, real-world business scenarios.Deep Decoding: How Large Language Models 'See' Structure?https://cuicaihao.github.io/posts/2026-05-03-how-llms-see-structure/Sun, 03 May 2026 09:05:02 +1000https://cuicaihao.github.io/posts/2026-05-03-how-llms-see-structure/Explore the architectural differences between Native Multimodal, MoE, and Long Context LLMs, and how they perceive and process structural information in professional workflows.The Great AI Dao Debate: A Xianxia Epic of Silicon Immortalityhttps://cuicaihao.github.io/posts/2026-05-01-ai-dao-debate-xianxia-epic/Fri, 01 May 2026 14:30:00 +1000https://cuicaihao.github.io/posts/2026-05-01-ai-dao-debate-xianxia-epic/A deep dive into the 2026 global AI landscape through the lens of Xianxia (Chinese fantasy). This epic narrative explores the strategic rivalries between major ‘sects’ like OpenAI, Meta, Google, Microsoft, Apple, DeepSeek, and Alibaba. It examines the fundamental ‘Dao’ debates: Scaling Law vs. World Models, and Open-Source vs. Closed-Source, mapping the path toward AGI in an era of silicon immortality.Breaking the Million-Token Barrier: 5 Impactful Takeaways from DeepSeek-V4https://cuicaihao.github.io/posts/2026-04-30-breaking-million-token-barrier-deepseek-v4-takeaways/Thu, 30 Apr 2026 08:55:59 +1000https://cuicaihao.github.io/posts/2026-04-30-breaking-million-token-barrier-deepseek-v4-takeaways/An analysis of DeepSeek-V4’s breakthroughs in context efficiency, architectural innovation, and stability, reducing KV cache by up to 93% and enabling million-token contexts.