Recent advancements in protein language models (pLMs) have shown significant potential in the field of protein design and engineering. One paper explores deep learning models that co-generate both protein sequences and structures, aiming to enhance the accuracy of protein design by modeling both modalities simultaneously. Another study investigates how pLMs use sequence likelihood to predict zero-shot fitness estimation, a crucial aspect of protein engineering and mutation prediction. Additionally, groundbreaking research led by Peter Kim and Brian Hie at StanfordEng and Stanford_ChEMH demonstrates that pLMs can predict allosteric residues without requiring structural data or multiple sequence alignments, highlighting the models' ability to learn allosteric interactions without explicit supervision. These advancements suggest that evaluating protein LMs solely on their ability to learn 3D structures may overlook important functional information. The research is available on bioRxiv.