A divergent thinking deck for LLM's! I like it. Have been experimenting with similar things, trying to build AI-based "innovation engines". Have you experimented with higher temperatures at critical points of the output generation and/or something like XTC (Exclude Top Choices) from llama.cpp?
Divergent thinking deck is a good way to put it! Haven't played with those recently since using a lot of frontier models where you don't really have access to those knobs unfortunately so have to mostly stick to prompting. This does have the advantage of being able to go beyond the information the model has internalized. Would imagine you could apply both to get way off the beaten path.
A divergent thinking deck for LLM's! I like it. Have been experimenting with similar things, trying to build AI-based "innovation engines". Have you experimented with higher temperatures at critical points of the output generation and/or something like XTC (Exclude Top Choices) from llama.cpp?
Divergent thinking deck is a good way to put it! Haven't played with those recently since using a lot of frontier models where you don't really have access to those knobs unfortunately so have to mostly stick to prompting. This does have the advantage of being able to go beyond the information the model has internalized. Would imagine you could apply both to get way off the beaten path.
[dead]