Divergent ideas, not safe ones. Pick Brainstorm from the sidebar when you want range, not a single best guess.
Brainstorm is the interceptor for when you want options. Pick it from the sidebar and the deep-thinking layer switches into a mode tuned to push for distinct, varied ideas instead of converging on the most plausible-sounding answer.
Most AI assistants quietly average toward the median. Brainstorm pushes the other direction: more spread, more variety, more rough edges. Useful precisely when the median answer is the wrong answer.
Brainstorm uses a high sampling temperature, which gives the model more freedom to take less-likely paths. That's what produces variety instead of a polished single answer.
On top of the model, Brainstorm layers a system prompt that asks for multiple distinct angles, with no early convergence. The framework is the proprietary part.
Brainstorm only kicks in on escalated turns. The fast layer stays out of it, so easy questions keep their snappy default behavior.
When you want a single, confident answer. Asking a Brainstorm-mode model "what is 17% of 240" will get you a correct answer with two unnecessary alternatives, which is not what you needed.
No. Higher randomness produces variety, and variety includes some misses. That's the point: you pick from a wider pool. If every idea were polished, the pool would have collapsed back to the median.
Yes. Switch back to no interceptor (or Reasoning) and ask the model to develop one of the ideas in depth. The conversation history carries over.