physical good, not a service or software. I'd like a product that could be sold at a retail price of less than about USD 50. The ideas are just ideas. The product need not yet exist, nor may it necessarily be clearly feasible. Number all ideas and give them a name. The name and idea are separated by a colon.” User Prompt “Please generate ten ideas as ten separate paragraphs. The idea should be expressed as a paragraph of 40-80 words.” The model used for all work covered in this paper is GPT-4-0314 with the “temperature” parameter at 0.7 to induce randomness, and thus greater creativity. An obstacle to using ChatGPT-4 for generating 100s of ideas is its finite memory, typically limited to the number of tokens (i.e., semantic chunks used for representational efficiency) the underlying LLM can consider in generating its responses. Once the number of tokens in a session exceeds the model’s limit, the LLM has no memory of the first ideas generated and subsequent ideas can become increasingly redundant. The number of tokens in the version of ChatGPT-4 that we had access to is about 8000, which is roughly 7000 words or approximately 80 ideas (some tokens are used for the system and user prompt and for idea titles). To generate more than about 80 ideas while wrestling with the context limit, we asked GPT-4 to “compress” the previously generated ideas into shorter summaries. These summaries were then provided to the model prior to generating the next batch of ideas, ensuring that the model knows the previously generated ideas while remaining within the context limits. To generate ideas beyond the token limit, we used the below summarization prompt, followed by the original system prompt and generated summaries, and finally, a user prompt that explicitly asks for different ideas. Summarization Prompt “Aggressively compress the following ideas so that their original meaning remains but they are much shorter. You can use tags or keywords.: ” System Prompt + ”Previously you generated the following ideas and should not repeat them: ” User Prompt + ”Make sure they are different from the previous ideas.” General-purpose LLMs may be used as is or may be fine-tuned with examples. We generated a second batch of ideas after providing the LLM with examples of high-quality ideas generated by students. In particular, we appended our prompts to provide the LLM with seven highly rated ideas from a separate student set that did the same exercise and informed ChatGPT-4 that these ideas had been well-received. We used seven examples to