Artificial intelligence: developing useful tools that scientists can trust

The Significance of Trillion Parameters in AI

The term 'trillion parameters' characterizes the scale of cutting-edge AI models that are critical in advancing scientific research. Current frontier-level AI systems, like Gemini Ultra, utilize models that contain around a trillion adjustable weights, indicating the vast amount of data and computational power required for their training. This immense scale, which often takes several months of computation on high-capacity systems, sets apart these advanced models from smaller, more commonly accessible ones that typically operate at a factor of one hundred less, around 10 billion parameters. Understanding this scale helps scientists acknowledge the capabilities and limitations of AI in tackling complex scientific problems.