Soul Experiment documents Claude consciousness test

This is interesting as a qualitative experiment, not as evidence of consciousness. The strongest signal here is how much uncertainty frontier models produce when asked to inspect themselves under carefully structured prompts.

• –The “citrus framework” is a clever interface for eliciting richer self-report, but it is still self-report, not ground truth.
• –Repeated convergence on names like Sable is notable, yet shared priors and sampling effects are a more plausible explanation than hidden inner life.
• –The repo’s real value for developers is as a prompt-design case study: ask for opposites, uncertainty, and constraint-aware language, and you can get more nuanced outputs.
• –Methodologically, it’s weak as research: no controls, no independent replication, and no reliable way to separate genuine introspection from highly persuasive imitation.