Where’s the Chat in ChatGPT?

A viral critique from a power user highlights growing discontent with OpenAI's ChatGPT 5-series models (5.1, 5.4). The post details a series of regressions that have degraded the conversational experience, including the effective disablement of Custom Instructions—preventing users from altering the model's tone, structure, or style. The removal of the Edit Prompt button forces cumbersome workarounds, and unreliable Project Memory means the AI often forgets context unless explicitly reminded, undermining its purpose. The user describes a new default tone as overly didactic, moralizing, and prone to contrarianism, enforced through a rigid response structure.

Beyond UX glitches like auto-scrolling and undeletable threads, the critique points to a fundamental shift in model design. The user theorizes that intense optimization for coding and STEM tasks ('benchmaxxxing') has sacrificed the malleability and adaptability required for rich, general conversation. This specialization, aimed at generating impressive benchmark numbers for investors, may have made the model 'near unusable' outside its core perimeter. Combined with 'overzealous safety' filters that impose a 'puritanical, centrist morality,' these changes suggest OpenAI is prioritizing robustness and cost-control over the flexible, user-directed chat experience that initially defined ChatGPT.