When Altman celebrates lastly getting GPT to keep away from em dashes, he’s actually celebrating that OpenAI has tuned the most recent model of GPT-5.1 (in all probability by reinforcement studying or fine-tuning) to weight customized directions extra closely in its likelihood calculations.
There’s an irony about management right here: Given the probabilistic nature of the difficulty, there’s no assure the difficulty will keep fastened. OpenAI constantly updates its fashions behind the scenes, even throughout the similar model quantity, adjusting outputs primarily based on consumer suggestions and new coaching runs. Every replace arrives with completely different output traits that may undo earlier behavioral tuning, a phenomenon researchers name the “alignment tax.”
Exactly tuning a neural community’s conduct isn’t but a precise science. Since all ideas encoded within the community are interconnected by values known as weights, adjusting one conduct can alter others in unintended methods. Repair em sprint overuse right now, and tomorrow’s replace (geared toward bettering, say, coding capabilities) would possibly inadvertently carry them again, not as a result of OpenAI needs them there, however as a result of that’s the character of making an attempt to steer a statistical system with hundreds of thousands of competing influences.
This will get to an implied query we talked about earlier. If controlling punctuation use continues to be a battle that may pop again up at any time, how far are we from AGI? We are able to’t know for positive, but it surely appears more and more possible that it received’t emerge from a big language mannequin alone. That’s as a result of AGI, a know-how that might replicate human common studying potential, would possible require true understanding and self-reflective intentional motion, not statistical sample matching that generally aligns with directions should you occur to get fortunate.
And talking of getting fortunate, some customers nonetheless aren’t having luck with controlling em sprint use exterior of the “customized directions” characteristic. Upon being informed in-chat to not use em dashes inside a chat, ChatGPT up to date a saved reminiscence and replied to 1 X consumer, “Obtained it—I’ll stick strictly to quick hyphens any longer.”
