Grammatical errors have different types, these include using wrong tenses or spelling errors. A person generally has a tendency to make the same type of errors subconsciously. By evaluating the type of error a person has already made, we can decrease ‘False Negatives’ knowing that these are errors the person is prone to make. Hence, Large Language Models (LLM) may be able to pick up these tendencies and adjust their outputs accordingly. Being able to pick up certain tendencies will allow us to adjust the model’s output to have more confidence in predicting the same type of tendencies. Our Seq2seq model has achieved this through “In-Context Learning (ICL)” and “Prefix Tuning”.