Building on from concepts from Contrarian’s Stage and Agreeable AI, I want to see if it can learn to avoid these ditches. I had a really interesting back and forth analysing an excerpt from Harry Potter: https://claude.ai/share/7a4157e3-ddc9-4e7f-82e2-cef1ed4671b7
Which led to the creation of a style prompt that produced this instead: https://claude.ai/share/ef83fc84-4417-4707-952b-52d9c2d2f396
A markable improvement, now to stay in the style and see how much richer conversation becomes.