If you ask LLM (ChatGPT-4o in this case), they can give even more example than you mention. Seems reasonable to me.
The article suggests that LLMs (Large Language Models) are capable of making discoveries beyond mere extrapolation, primarily by applying logical inference to new or neglected data. Here are some additional ways LLMs might contribute to new discoveries:
Pattern Recognition in Complex Data: LLMs can identify patterns in large and complex datasets that humans may overlook. These patterns can lead to the identification of previously unknown relationships or phenomena, which could be considered a new discovery.
Hypothesis Generation: LLMs can generate plausible hypotheses by combining information from diverse sources, allowing for the exploration of ideas that haven't been fully explored or considered by researchers.
Data Mining and Correlation Analysis: By processing vast amounts of structured and unstructured data, LLMs can discover correlations and connections that may lead to new insights or discoveries in various fields like science, medicine, and economics.
Creative Problem-Solving through Novel Combinations: LLMs can synthesize information across disciplines, forming connections between seemingly unrelated fields. This interdisciplinary approach can result in innovative solutions or discoveries that might not emerge from traditional human research methods.
Automated Literature Review and Summarization: LLMs can efficiently analyze and summarize large bodies of scientific literature, identifying gaps or trends that could lead to new research directions and potential discoveries.
Simulation of Unconventional Scenarios: LLMs can simulate novel scenarios or conditions that have not been considered, which might inspire new ways of thinking about a problem and lead to breakthroughs in fields like physics, engineering, or biology.
Assistance in Experimental Design: LLMs can help in designing experiments or research methodologies that might test novel hypotheses or theories in new ways, potentially leading to unexpected discoveries.
Reanalysis of Historical Data: LLMs can go through old or neglected data, reinterpreting it in light of modern theories and technologies. This could reveal new insights or discoveries that were previously missed due to limitations in past methods of analysis.
Combinatorial Innovation: LLMs can mix and recombine existing ideas, technologies, or data in unique ways, sometimes resulting in breakthroughs that are more than the sum of their parts. This form of "combinatorial creativity" can lead to novel discoveries.
Refinement and Optimization of Existing Theories: While not discovering entirely new paradigms, LLMs can refine existing theories by identifying inconsistencies or suggesting improvements, leading to a deeper understanding or more precise applications of those theories.
In sum, while LLMs may not have intuition or creativity in the human sense, they can still engage in processes that lead to discoveries, particularly through their ability to handle vast amounts of information, identify patterns, and apply logic.
