We wonder—how many SoSyM readers have tried using OpenAI’sChatGPT to formulate prompts to explore the interesting results that may be produced? It is really fascinating to consider the results when the query is defined appropriately. Of course, we abstained from using ChatGPT to write any part of this editorial, but if we actually would generate portions of our text, we wonder how difficult it would be for readers to identify the generated parts. ChatGPT is rather chatty, but even the produced text can be adapted by asking it for a specific answering style. The options and future extensions seem to be endless. In the context of this editorial, we do not consider legal or ethical issues of ChatGPT—these have to be sorted out eventually. We also do not discuss the problems of having ChatGPT text, or text generated by other Large Language Models (LLMs), in scientific papers. Rather, in this editorial, we focus more positively and ask the main question: How will ChatGPT be able to help us in a development process, especially in the tasks of developing or using models for analysis, production or understanding of software and systems? ChatGPT can provide useful pieces of programs that are sometimes correct and perform the task that the user asked to be implemented. ChatGPT also knows several textual modeling languages. This even seems to hold for DSLs. The example of GraphGPT suggests that a language designer can tell ChatGPT what the desired modeling language looks like and it will produce pieces within that language. GraphGPT uses the trick of asking for a JSON-encoding of the graph that is rendered into a diagram. The options seem endless
No takes yet. Share an insight, caveat, or question.
Combemale et al. (2023) studied this question.