Comparing Models
These are several models suitable for brainstorming-related tasks. The LM Arena's "Hard Prompt" category ranks models based on prompts that include problem-solving, ambiguity, creativity and reasoning which makes it a good proxy to find models for brainstorming and ideation. (visit the leaderboard)
As of October 2024, OpenAI's models are the best.
3rd place models are still strong contenders and may be superior in many situations. For example, Claude 3.5 Sonnet was selected as "better" than gpt-01-preview for approximately one of every three questions.
Tip #1: Use ChatGPT for general brainstorming
The model that works best for a given request may change and all the top consumer apps have good models. Because of that, I recommend using multiple applications, but finding the one that works best for you in terms of both the response quality, features and user experience.
Beyond Basic ChatGPT
Effective brainstorming requires finding solutions that adhere to a set of constraints. Unearthing those constraints over and over again can leave your fingers tired - but there are other tools you can use as well as features directly within your favorite conversational agents.
Tip #2: Take advantage of Claude Projects
AI assistants can index and search through files in what is called Retrieval-Augmented-Generation (RAG). As a consumer you can leverage Claude Projects, ChatGPT Custom GPTs and Google's NotebookLM to upload pdfs and documents. This means that instead of starting a conversation by providing the same context over and over again you can create a project (or custom GPT) that stores and remembers this information.
As an example I created a Claude Project with information about a fundraising effort that I'm leading. I added a document that showed the fundraising activities we've done to-date, and could then brainstorm additional activities, without having to repeat myself, if I start new conversations. I like the user experience of saving and publishing artifacts and viewing historical conversations within a project so prefer Claude Projects over NotebookLM and Custom GPTs.
This is really powerful for Enterprises when you want to brainstorm and ideate but want the LLM to have the context of current projects, business priorities, policies, marketing efforts, etc. For that reason we now see organizations building internal AI assistants that index company policies and documents, and products like Glean that offer this as an off-the-shelf solution.
Tip #3: Talk through it with Advanced Voice Mode
Advanced Voice Mode is now broadly available for ChatGPT and Microsoft CoPilot users. If you're an auditory processor and like to think through problems out loud this is a great tool for brainstorming. It goes beyond the single question-answer conversations offered by Siri and Alexa and instead allows you to interrupt and converse in a more natural way. The feature is only available in mobile apps.
Are you brainstorming with a group or more of a visual learner?
Products like Ideamap.ai take digital brainstorming tools to the next level. If you've ever created post-its in Miro and then needed to extract themes and group them - that's exactly the type of features you'll find in Ideamap.
Perfecting your prompts
Tip #4: Iterative Feedback Loops
Engage in a dialogue with the AI to clarify and refine its suggestions. Ask follow-up questions, request elaborations, or challenge assumptions to push the ideation process further. LLMs are great for divergent thinking; people come in for convergent thinking.
Question the AI's design choices and ask it to justify or simplify its suggestions. This helps avoid unnecessary complexity and ensures ideas remain practical and feasible.
See an example
[…]
Tip #5: Structured Prompts
Craft specific, well-structured prompts (like Action-Objective-Context) to guide the AI's responses. Clear instructions and constraints can result in more relevant and focused output, enhancing the efficiency of your brainstorming session.
See an example
Tip 6: Quantity Over Quality Initially
Begin by generating a large number of ideas. This approach capitalizes on the AI's ability to produce diverse concepts rapidly, providing a rich foundation for further refinement. Then craft prompts that evaluate ideas for quality, feasibility, etc. and refine into a more manageable list you can work with.
See an example
Tip 7: Be explicit in instructions.
Without explicit instructions to "consider diverse approaches" to avoid fixation you may find that ideas generated by an LLM are repetitive. This prompt as well as giving explicit instructions to come from a different angle can help stimulate creativity in responses.
Similarly, triggering a question and answer mode, where you instruct an LLM to "ask me reflective questions to clarify" will encourage back-and-forth that leads to better brainstorming. On its own the LLM will provide answers with assumptions rather than ask you to provide more information, so this can make a big difference in brainstorming.
See an example
[…]
[…]
Tip 8: Break down complex asks
Rather than directly asking an LLM to brainstorm the output you're interested in, break that ask into smaller less complicatd asks, and tackle them individually, just as you would when doing the work on your own. This is very similar to recommendations for writing, where you would start by researching a topic, organizing and finding themes, and then building an outline — all before writing a final piece. If you're not sure how to break your ask down into more simple sub-problems prompt AI to identify tasks needed; start with the first task and work through each.
See an example
Instead of asking Claude to help you come up with an itinerary for an upcoming vacation you might try
- a Q&A brainstorm to come up with criteria for what makes a great vacation
- Selecting a general location
- Selecting the several possible things to do across several categories, within the location selected and criteria.
- Drafting 3-5 possible itinerary based on the criteria and constraints for what you can do within the time period of the vacation, costs, etc.
- Finalizing an itinerary
Tip 9: Use multiple LLMs
Research shows the greatest diversity comes from ideas combined from multiple LLMs. So when you're trying to maximize for creativity use 2-3 LLMs and then combine results.
Tip: Use poe.com or another LLM aggregator to talk to multiple LLMs in a single conversation for a seamless user experience.