Raw LLM Responses

The answer is data availability We have plenty of open source code online that can be used for training deep learning models At first, it was just for the sake of research, but obviously greedy people thought: "I can use it to replace programmers" The same happens for music, movies and illustrations Also the scientists that trains AI knows how to code, therefore they can easily verify for correct answers (and can execute the code for assessing the result). Since AI is very good at fooling people in a convincing way, it would be hard for data scientists to evaluate a "lawyer agent" at scale.