A study of 82 people found that using generative AI for a creative task left the actual output quality basically unchanged. What dropped was how much people trusted their own creative ability afterward. Separate research on heavy language model use found a similar pattern at scale: more homogenous ideas, fewer genuinely odd ones.

None of this is an argument against using the tools. dreamsedge runs on AI too. But handing someone an answer and helping them find one on their own are different processes, even when they produce a similar looking result, and staying on the right side of that difference is something dreamsedge has to keep earning, not something it gets to assume.

The AI inside dreamsedge is built to ask better questions about what you actually want, not to answer them for you. If it ever started doing the opposite, it would be solving the same wrong problem this study just measured.

Dreaming, it turns out, works the other way. A recent study found people whose sleep was nudged toward an unsolved problem woke up having actually solved it more often than people left alone. Nobody handed them the answer. They just weren't trying so hard while their mind kept working.

Borrowing an idea and generating one aren't the same use of a mind. dreamsedge runs on the first and is trying to build toward the second, which is a harder thing to promise but the only one worth building toward.