Major language models (llMS) can be your AI agents's treasure in all domains, as long as he or she can accurately follow the user's orders. However, the latest courses show a major limitations in the LLMS history skills, raise concern about their reliability of maximum applications. Accurate equality of llms' uncertainty to the instructions are important in reducing the risk of submitting. We introduce, our knowledge, the first formal assessment of uncertainty skills of llms in the teaching context – the following. Our research shows important challenges with existing benches that follow the teaching, where many things are included and uncertain about the current instruction, compared to the separation of other ways and models. Dealing with these issues, introduces an assessment management of the two Bemchmark, which enables the complete comparisons of uncertain measures under different conditions. Our acquisition shows that the unethical measures fight, especially when models make brilliant mistakes in the following instructions. While the internal model states that it provides some development, they remain enough in the most complex situations. Understanding from our controlled controlled care provides valuable understanding of the llms and the existence of measurement-based on the instructions, expressing AI independent lawyers of AI
