Reflections on AI at the end of 2025

The blog post reflects on advancements in AI by 2025, specifically addressing the evolution of Large Language Models (LLMs) and the significance of Chain of Thought (CoT) in improving model outputs. It highlights the changing perceptions of AI researchers on LLM capabilities, the impact of reinforcement learning, and the adoption of AI tools by programmers. The author argues that while some old views on LLM limitations have shifted, the core architecture remains the same, and anticipates that AGI could emerge through various architectures in the coming years, along with addressing existential threats in AI development.