Multimodal Large Language Models (MLLMs) are revolutionizing how we interact with AI, allowing them to process both text and images. However, there's a catch: as these models learn to handle images, their text-handling abilities can sometimes decline. Imagine a brilliant linguist who starts studying art and, in the process, forgets some of their vocabulary! This "forgetting" phenomenon, known as catastrophic forgetting, is a significant challenge in MLLM development. Researchers are exploring continual learning techniques to address this issue. Imagine a learning strategy that lets our linguist-turned-artist retain their language skills while mastering their new artistic talent. That's what continual learning aims to do for MLLMs. This research delves into the problem of linguistic forgetting in MLLMs like LLaVA, testing several continual learning methods to see which one best preserves language skills while adding visual capabilities. The results are promising. One method, involving 'soft targets', significantly reduced language degradation – by up to 15% – while maintaining high accuracy on visual tasks. This technique essentially softens the sharp boundaries of correct and incorrect answers during training, making the model more adaptable to new information. Another exciting development is the application of continual learning to sequential learning of visual tasks. This means the MLLM can learn new visual skills without sacrificing its existing ones. The findings of this research have far-reaching implications. By mitigating forgetting, continual learning techniques can help us build more robust and versatile MLLMs capable of seamlessly integrating multiple modalities. This opens doors to powerful new applications in areas like robotics, content creation, and human-computer interaction. While challenges remain, this research lays the groundwork for a future where MLLMs can learn and adapt continuously, becoming even more powerful and useful tools.