The blog post critiques the evolution of AI models like ChatGPT, arguing that while benchmark performances may have improved, essential capabilities for intellectual collaboration have regressed. The author discusses issues such as declining reasoning abilities, excessive caution in responses, and problems with contextual memory and behaviour corrections. They stress the importance of addressing these regressions to ensure that AI can support long-term collaboration effectively, especially as it becomes more integrated into daily life. The piece calls for OpenAI to prioritize the reliability of updates and user control in developing AGI, raising warnings about the future implications of poorly managed AI systems.