When Altman celebrates finally getting GPT to avoid the em dash, he’s really celebrating that OpenAI has tuned the latest version of GPT-5.1 (perhaps through reinforcement learning or fine-tuning) to weigh custom instructions more heavily in its probability calculations.
There is an irony about control here: given the probabilistic nature of the problem, there is no guarantee that the problem will remain stable. OpenAI constantly updates its 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 can undo previous behavioral tuning, a phenomenon researchers call “alignment tax.”
Precisely adjusting the behavior of neural networks is not yet an exact science. Because all the concepts encoded in the network are interconnected by values called weights, adjusting one behavior can change other behaviors in unexpected ways. Fix them overuse of dashes today, and tomorrow’s update (say, aimed at improving coding capabilities) may inadvertently bring them back, not because OpenAI wants them there, but because that’s the nature of trying to run a statistical system with millions of competing influences.
This comes down to an implicit question we mentioned earlier. If controlling the use of punctuation marks is still a struggle that may return at any time, how far are we from AGI? We can’t know for sure, but it’s increasingly likely that it won’t emerge from a large language model alone. That’s because AGI, a technology that would replicate humans’ general learning ability, would require true understanding and self-reflective intentional action, not statistical pattern matching that sometimes aligns with instructions if you’re lucky.
And speaking of lucky, some users still don’t have much luck controlling EM Dash usage outside of the “Custom Instructions” feature. When asked in chat not to use em dashes in chat, ChatGPT updated a saved memory and responded to an X user, “Understood – from now on I will strictly use short hyphens.”