The Heidelberg-based start-up has developed a new AI ... LLM. The problem arises when the new text differs greatly from the text used to train the LLM during fine-tuning. Then, as Aleph Alpha ...