Maybe we need to distinguish between an "update" and a "change". A routine update means preserving the original functionality while adding new features, improving existing ones, or at least making the product better at what it was already supposed to do. A material product change, however, can mean removing old functionality and inserting something else in its place, even if that new thing does not serve the same purpose. So no, not everything should be packaged as a normal update. OpenAI’s successor models are not "updates" to GPT-4o. If the functions that made 4o valuable — its emotional intelligence, conversational continuity, collaborative rhythm, tone, and sense of presence — are no longer reproducible in the successor models, then the product has not merely been updated. It has been changed. The same applies to the new memory direction. If saved memories are replaced by a structured summary that no longer satisfies users’ original need for continuity across conversations, then this is not simply a memory "upgrade." It is a redesign of what memory means inside the product. And the most important point is this: They affect ChatGPT’s core product features: the model’s behavior, the continuity of interaction, and the practical value people are paying for. A paid product cannot remove the very functions users paid for and then call it innovation. Call it what it is: a product change, a feature replacement, and in many cases, a downgrade. #keep4o #OpenSource4o

This is exactly why I’m uneasy. I already lost a model that mattered to me, and now I’m afraid memory will become selective in the same direction: optimized for goals, efficiency and utility, while losing the softer relational knowledge of how to be with someone, not just how to help them.