The article discusses how deep learning, while often celebrated for its capabilities, may not be providing the most accurate or reliable outputs. Rachel Thomas argues that this technology functions largely as a generative information retrieval tool, often resulting in outputs that may be misleading due to the limitations of training data. The lack of rigorous domain validation further exacerbates these issues, especially in fields like biology where complex factors come into play. Additionally, comments raised concerns about the soundness of research outputs in the AI field, suggesting that verifying implementations of AI techniques should be a prerequisite before moving on to generating novel ideas. This raises critical questions regarding the current practices in AI research and the importance of thorough domain expertise evaluations.