6 The ruggedness of the underlying solution landscape thereby makes it impossible to arrive at the most valuable idea in the idea space via planning and synthesis. Rather, a broad exploration is needed which calls for the generation of multiple ideas of which the idea(s) with the highest value will be retained while the other ideas are likely going to be discarded. The success of the exploratory effort is measured by the value of the most valuable idea that was generated. To make the exploration of the solution landscape effective, it is important to consider a wide array of ideas, each corresponding to a different point in the landscape. To find the most valuable idea thus requires the generation of very different starting points. We refer to this as the diversity of a given pool of ideas. A pool of ideas is diverse if it includes ideas with very different coordinates in the landscape. This can be easily visualized in the two dimensional space of Figure 1. As mentioned above, a more diverse pool of ideas is thus likely to come up with a better value of the best idea. Based on prior work, we hypothesize that AI-based idea generation will lead to less idea diversity. Yet, we also hypothesize that idea diversity can be increased via prompt engineering, with CoT prompting being yielding the most diverse ideas. Specifically, we state: H1: A pool of AI-generated ideas with no specific prompting is less diverse than a pool of ideas generated by groups of humans. H2: The prompts used in generating a pool of ideas using AI can increase the diversity of the ideas. H3: Chain-of-Thought prompting will be the most effective in increasing the diversity of the ideas. It might or might not outperform human group idea generation. Prompting Strategies for Innovation and Brainstorming As a baseline for comparison, we use a pool consisting of 100 randomly chosen ideas generated by Wharton MBA students. This is the same pool of ideas used as in Girotra et al 2023. It is important to note that students were not tasked with generating 100 ideas at once, but instead the pool is an aggregation of ideas from multiple students. We want to explore how different ways of prompting the LLM might impact the diversity of the resulting idea pools. In total, we evaluate 35 prompting strategies.