When Altman celebrates finally getting GPT to avoid em dashes, he's actually celebrating that OpenAI has tweaked the latest version of GPT-5.1 (likely through reinforcement learning or fine-tuning) to take user instructions more heavily into its probabilistic calculations.
There is an irony here about control: given the probabilistic nature of the problem, there is no guarantee that the problem will remain solved. OpenAI constantly updated their models behind the scenes, even within the same version number, adjusting the output based on user feedback and new training runs. Each update comes with different output characteristics that may override the previous behavioral setting. Researchers call this phenomenon “equalization tax“
Fine-tuning the behavior of a neural network is not yet an exact science. Because all concepts encoded in the network are interconnected values called weightsadjusting one behavior may unintentionally change others. Fix em Dash's overuse today, and tomorrow's update (aimed at, say, improving coding capabilities) might inadvertently bring them back, not because OpenAI wants them there, but because that's the nature of trying to manage a statistical system with millions of competing influences.
This leads to the implied question we mentioned earlier. If controlling the use of punctuation is still an issue that could arise again at any time, how far are we from AGI? We can't know for sure, but it seems increasingly likely that it will emerge from more than just a large language model. That's because AGI, a technology that replicates general human learning abilities, will likely require true understanding and self-reflective goal-directed action rather than statistical pattern matching that sometimes matches instructions if you're lucky.
Speaking of luck, some users still have trouble controlling the use of the em dash outside of the “user instructions” feature. When told in chat not to use em dashes, ChatGPT updated the stored memory and replied to one user X: “I see, I'll be strictly sticking to short hyphens from now on.”
