The US Used to Demand the Best Tech. Now We Ban It
Summary
A reflective exploration of testing whether large language models can recognize and handle alterations to their own outputs. The author argues that traditional mirror tests probe the wrong capability, using an olfactory analogy for self-recognition and conducting informal experiments across Gemma, GLM, and Claude to observe how models react to deliberate corruption of their own text. The piece questions whether AI self-awareness is plausible or simply emergent from training on human-like patterns, and it calls for more rigorous, controlled experiments.