OpenAI CEO Sam Altman has argued that AI models should eventually produce synthetic data good enough to train themselves effectively. As the well of usable human-generated data dries up, more companies look into using synthetic data. Rather than being pulled from the real world, synthetic data is generated by AI systems that have been trained on real-world data. Synthetic data may help offer some effective "countertuning" to the biases produced by real-world data, too. 'Habsburg AI'While the AI industry found some advantages in synthetic data, it faces serious issues it can't afford to ignore, such as fears synthetic data can wreck AI models.

---

Persons: —, that's, Sam Altman, Gary Marcus, It's, Nathan Lambert, Gretel, SynthLabs, Meta, Timnit Gebru, Margaret Mitchell, LLMs, Sadowski, Alexandr Wang, AlphaGeometry, Marcus Organizations: Service, Google, Business, Oxford, Gartner, New York University, Allen Institute, AI, Nvidia, Meta's, Anadolu, Getty, Rush, Microsoft, Monash University Locations: Cambridge, Habsburg