This may be obvious to many people already, but I recently thought about this in a few scenarios and figured it may be valuable to articulate it clearly. One note before the main discussion is we should acknowledge that LLMs do have some sort of independent and creative thinking. Some folks are still skeptical about…

Hacker News is a community-driven platform for sharing and discussing technology news, startups, and programming-related topics. Through user submissions and comments, Hacker News offers insights into emerging technology trends, industry developments, and entrepreneurial ventures. Readers can participate in discussions, share their insights, and stay informed about the latest advancements in technology and innovation.

Hacker News

LLMs are not a good tool for effective brainstorming due to their training to follow existing patterns and converge to existing consensus. Potential solutions to improve LLM training for brainstorming are discussed.

LLMs are not suitable for brainstorming

<p>Both sides of this are describing different halves of brainstorming. Divergence, getting the twenty obvious options on the table fast, is something a model does better than a room of tired people. Genuine novelty is a different ask, and convergence toward consensus is not a bug you can prompt away - it is the objective the model was trained on. Where I notice the ceiling is that the ideas cluster: ask for ten approaches and you get one idea in ten costumes, because they all sit in the same neighbourhood of the training distribution. The practical fix is constraint rather than better prompting - forbid the obvious answer, fix a hostile budget, require it to argue for the option it just rejected. Novelty comes from the constraint, not from the model.</p>


<p>that’s a wild, hard to back statement! given how successful brainstorming with LLMs are, I think this needs to be qualified a lot better, there is no other technical tool that I know of that has been more effective to brainstorming in the history of technology, so I’d infer that, this is intended as we can have even better tools than LLMs for brainstorming</p>
