Maximizing AI Model Size and Computing Power

The key to stealing an AI model is obtaining the connections or weights that encode all the knowledge within the system. The more massive the system, such as GPT-4 with around a trillion connections, the more complex it is to make it work coherently. Scaling up the model size requires a proportional increase in computing power for tuning. Increasing both the model size and computing power simultaneously results in significant improvement in performance, essentially leading to a higher IQ for the AI system. This realization in 2020 has prompted the AI industry to focus on maximizing the equation between model size and computing power to enhance AI efficiency.