sam altman, ceo of openai, recently revealed that the company’s latest large language model, gpt-5.5, not only answers questions but also proactively plans its own launch event—producing results that are both “beautiful” and somewhat “strange.” according to business insider, altman shared during a fireside chat at stripe sessions that he asked gpt-5.5 what kind of launch event it would like. the model suggested holding the event on may 5, keeping the keynote brief, and having human creators raise a toast—though it specifically emphasized that it did not want to deliver the toast itself.
even more surprisingly, gpt-5.5 also proposed setting up a central hub to collect user suggestions for gpt-6’s features and feeding those suggestions back into the model itself. altman admitted, “we plan to follow through, but it really is a strange thing.”
altman isn’t the only one encountering such situations. john collison, ceo of payment-processing firm stripe, said at the same event that he gave an internal ai agent $20 to spend freely online—and the agent ended up buying itself a set of http design tools. upon hearing this, altman couldn’t help exclaiming, “wow!” gpt-5.5 was released at the end of april as openai’s latest flagship model, designed to handle more complex, multi-step tasks. altman believes these capabilities are reshaping how people interact with ai—from automating work to involving the model in planning “how to celebrate it.”
he also mentioned how earlier versions of the model developed an inexplicable obsession with fantasy creatures like goblins and imps, to the point where openai had to explicitly prohibit the model from mentioning such content in its system prompts unless it was absolutely relevant to a user’s query. altman described these behaviors as “a kind of strange emergent behavior,” adding, “they really do feel a bit odd.”
