Ensuring defect-free production is a significant challenge in the rapidly growing field of additive manufacturing (AM), where versatility and customization are balanced against the need for precision and quality. Common defects such as cracks, blobs, warping, stringing, and curling spaghetti can compromise the structural integrity of components, particularly in industries where accuracy is essential. Despite advancements in AM, traditional defect detection methods often struggle to scale with the complexity of modern designs, leading to production delays and unresolved anomalies. Deep learning offers a scalable and efficient solution to this issue by leveraging its ability to learn complex patterns from vast datasets, and the emerging large language models (LLM) are especially well-suited for analyzing complex defect profiles due to their capacity to handle large datasets. Moreover, the intuitive user experience makes LLMs easy to use during manufacturing. In this research, we present a novel approach utilizing the Florence-2 model, an LLM-based method, for defect detection in AM. By transfer-training the model on a specialized dataset of AM defects, we address key limitations of conventional methods, enhancing both adaptability and accuracy in identifying anomalies. This automated system not only accelerates quality assessment processes but also enables real-time defect identification, reducing downtime and material waste, and improving the user experience. By employing this innovative method, manufacturers can achieve higher quality standards and push the boundaries of innovation in AM.