The author reports a recurring reliability issue with ChatGPT, detailing two main failure modes: abrupt termination of the thinking process without generating answers and incomplete responses during generation. These failures occur frequently (30-50% of requests), severely affecting the usability of ChatGPT for professional and research tasks. The post emphasizes a need for OpenAI's technical team to investigate these issues, check server-side logs, and provide resolution timelines.