The field of natural language processing and artificial intelligence is moving towards developing more robust and responsible language models, particularly in domains such as law and chemistry. Researchers are focusing on creating benchmarks and evaluation metrics to assess the reasoning capabilities of generative language models, highlighting their limitations and brittleness. Meanwhile, innovative approaches are being proposed to extract chemical data from scientific literature, including vision-based deep learning frameworks and curated question-answer databases. These developments have the potential to improve the efficiency and accuracy of chemical research and applications. Noteworthy papers include:

• Parameterized Argumentation-based Reasoning Tasks for Benchmarking Generative Language Models, which introduces a novel approach for evaluating reasoning capabilities of language models.
• MolMole, a vision-based deep learning framework for extracting molecular structures and reaction data from scientific documents.
• Enigme, an open-source library for generating text-based puzzles to evaluate reasoning skills in language models.
• ChemRxivQuest, a curated chemistry question-answer database extracted from ChemRxiv preprints.