The Arms Race in AI: Finding Non-Obvious Sources of Training Data

The value in AI models lies in finding non-obvious sources of data to improve training. While models like chat GPT can easily handle simple questions and provide correct answers, the real challenge lies in achieving precision and accuracy for more complex and important tasks. Currently, these models have not reached the level of being overwhelmingly right in such cases. However, once this hurdle is overcome, AI models will become commoditized as a service. The real competition will be in finding proprietary sources of data to train on. For example, a company that aims to detect tumors may gain an advantage by vertically integrating and acquiring a hospital system to access proprietary patient data. In the next five to ten years, we can expect to see exciting moves like these in the business world, which will make AI more useful and impactful.