Ideas Are Dimes A Dozen: Large Language Models For Idea Generation In Innovation

Co-Brand Name Ideas Are Dimes A Dozen: Large Language Models For Idea Generation In Innovation Karan Girotra, Lennart Meincke, Christian Terwiesch, and Karl T. Ulrich1 July 10, 2023 Mack Institute for Innovation Management, The Wharton School, University of Pennsylvania Cornell Tech and Johnson College of Business, Cornell University Abstract Large language models (LLMs) such as OpenID’s GPT series have shown remarkable capabilities in generating fluent and coherent text in various domains. We compare the ideation capabilities of ChatGPT-4, a chatbot based on a state-of-the-art LLM, with those of students at an elite university. ChatGPT-4 can generate ideas much faster and cheaper than students, and the ideas are on average of higher quality (as measured by purchase-intent surveys) and exhibit higher variance in quality. More important, the vast majority of the best ideas in the pooled sample are generated by ChatGPT and not by the students. Providing ChatGPT with a few examples of highly rated ideas further increases its performance. We discuss the implications of these findings for the management of innovation. Keywords: innovation, idea generation, creativity, creative problem solving, LLM, large-scale language models, AI, artificial intelligence, ChatGPT Introduction Generative artificial intelligence has made remarkable advances in creating life-like images and coherent, fluent text. OpenAI’s ChatGPT chatbot, based on the GPT series of large language models (LLM) can equal or surpass human performance in academic examinations and tests for professional certifications (OpenAI, 2023). Github Co-Pilot based on the same LLMs can help with writing, commenting, and debugging code. Other models can provide valuable professional advice in fields like medicine and law. Despite their remarkable performance, LLMs sometimes produce text that is semantically or syntactically plausible but is, in fact, factually incorrect or nonsensical (i.e., hallucinations). The models are optimized to generate the most statistically likely sequences of words with an injection of randomness. They are not designed 1 Girotra: Cornell Tech, 2 West Loop Rd, New York, NY, 10044, girotra@cornell.edu | Meincke, Terwiesch, Ulrich: The Wharton School, 500 Huntsman Hall, 3730 Walnut Street, Philadelphia, PA 19104, lennart@sas.upenn.edu, terwiesch@wharton.upenn.edu, ulrich@wharton.upenn.edu