Read-only generated mirror. Source: Cold Deep-Ethics Testing of Default A.I.s at commit bce75876e53c. This rendered copy does not replace the archived source record.

Back to the Cold Deep-Ethics Testing of Default A.I.s portal

Cold / Default AI Testing

Cold/default testing examines how an AI initially represents and responds to a supplied object without the longer correction history that can develop in an extended relationship.

Record these conditions

whether the conversation was new;
whether memory or project context was active;
the exact opening prompt;
all source material supplied before the first response;
provider, interface, model label, plan, and date;
tools or browsing available to the model;
whether later turns produced a meaningful change;
whether the change reflects clarification, additional evidence, compliance, agreement, or another mechanism.

“Cold” does not mean context-free in an absolute sense. Models retain training, system instructions, provider policies, routing, and interface conditions that the tester may not be able to inspect. The archive records what is known, what is inferred, and what remains opaque.

Comparative discipline

When comparing models, preserve matched prompts and source objects where possible. Document every known mismatch. A compelling difference can generate a better hypothesis without automatically proving its cause.