OpenAI CEO Sam Altman celebrated a small victory on X: ChatGPT now follows custom instructions to avoid em-dashes, a tweak that some users have long awaited.
The update lands just days after OpenAI released the GPT-5.1 model, and it has sparked mixed reactions: some see it as a sign of improved instruction-following, while others warn that a punctuation tweak is a narrow test of capability and does not prove the arrival of artificial general intelligence (AGI).
Several users weighed in online, with responses ranging from cautious optimism to skeptical takes that control over a single punctuation rule is a far cry from reliable general intelligence.
From a technical standpoint, instruction-following in large language models is not deterministic. Custom instructions influence token selection by nudging probabilities, but the model still weighs vast training data and current prompts, meaning a simple directive like “avoid em-dashes” can be ignored in some contexts.
While Altman\’s post underscores a potential step forward, experts say true AGI will require deeper understanding and autonomy beyond pattern-matching language models, and OpenAI continues to update its systems through RLHF and ongoing fine-tuning, which may reintroduce or remove behaviors over time.
