Recently I asked deepseek about how to use javascript to extract and makes computations with moodle in education. I note that the program did not consider two crucial points: 1) It is of utmost importance that the answer and the grade of the answer should be in the same row. 2) Don't modify the answer of the student. In this case the answer are ten letters in response to a test with ten questions. Deepseek used html closest to look for matching elements what is highly insecure, also it did pack answers with less that 10 letters with "a" what is a ridicule measure. This trivial problem reveals that deepseek is not able to consider the importance of the two trivials facts and that you have to introduce those two crucials points to avoid code that can mix students answers and grades. I think this illustrates clearly the current limitations of deepseek and the kind of context you have to give the LLM to avoid it. It seems that deepseek while coding was not alert to the kind of errors that could happen and their consequences. So you, the user of deepseek must provide this LLM with that context, you must think about the most simple program the LLM is about to design and how you must alert it to have special care to avoid failing completely, that is in the most destructive way.

1) answerCells.forEach(cell => { const row = cell.closest("tr");

2) return res.padStart(10, "a");