Sam Altman, CEO of OpenAI, has humorously celebrated a small but notable milestone: ChatGPT now correctly follows em dash formatting rules. In a lighthearted post, Altman noted that the model’s ability to properly use em dashes—those long, connecting punctuation marks used to set off parenthetical thoughts—marks a meaningful step forward in instruction-following precision. The moment, while seemingly minor, underscores a persistent challenge in AI development: even advanced models often struggle with nuanced, context-sensitive language rules. Despite significant progress in natural language understanding, AI systems frequently misinterpret or ignore subtle formatting instructions, leading to outputs that are technically correct in meaning but flawed in style. Altman’s tweet highlighted that this achievement is not just about punctuation—it’s a sign that models are becoming better at parsing complex, implicit directions. “We’ve been training models to follow instructions for years,” he wrote. “Today, they finally got the em dash right. Small win, big progress.” Experts agree that such milestones, while modest, reflect incremental improvements in model alignment and reasoning. The ability to correctly apply formatting rules requires more than just grammar knowledge—it demands an understanding of stylistic intent, context, and editorial judgment, all of which are hallmarks of human-level language proficiency. Still, the episode also serves as a reminder that true artificial general intelligence (AGI) remains far off. While models like ChatGPT can now handle em dashes, they continue to falter on more complex tasks, such as consistent long-form reasoning, deep causal understanding, or reliable self-correction. The fact that a CEO is celebrating a punctuation fix underscores how far AI still has to go to match human-level consistency and nuance. The moment has sparked both amusement and reflection across the tech community. Some see it as a sign of progress; others use it to question the hype around AI’s current capabilities. Either way, it’s a reminder that even the most advanced systems are still learning the basics—like when to use a dash, and when to leave it out.