I'm calling the ChatGPT from Python. My goal is to generate store is in a foreign langauge (Hebrew) that use only the vocabulary that the students have learned so far (by Lesson 12 of the textbook). I made Anki flashcards for the students with the help of ChatGPT, so I just reformatted those into a list of Hebrew words.
This is my prompt:
prompt = "Write a coherent 5 paragraph story (using only present tense verbs) in the Hebrew language." + \
"The story should be for beginner Hebrew students, but should be cohesive and interesting." + \
"After that, write the English translation of the story. The Hebrew letters should be written with vowels." + \
" For the Hebrew story, Use ONLY the Hebrew words listed below (and conjugations and variations of them); " + \
"however if you use any words not in this list, please list those words before the story with their English meaning: " + \
vocabulary_words
Issues:
- ChatGPT ignores this instruction: "Use ONLY the Hebrew words listed below". It typically uses 5 or 6 words not in my vocabulary list.
- Because of the issue above, as a backup plan, I asked it to list words that are no in my vocabulary list at the top of the story. It will typically do 2 or 3 words, but not all of them.
- Sometimes it makes nonsense sentences. This is an example of an English translation of one sentence: "People are also happy when they see that the library is not alone."
Is there anything I can do to improve my prompt to solve these issues? I might try Claude later to see how well it does. Would any other LLM be better at this.
