Homologous Blind Spot
A cognitive bias concept in AI-assisted development scenarios: when code generators and reviewers share similar model weights, training data distributions and prompt structures, their sensitivity to errors overlaps heavily, causing systemic problems to be jointly overlooked. It differs from blind spots arising from the limited experience of human reviewers.
Timeline (last 90 days)
Another camp of commenters worries about "same-origin blind spots": if the reviewer and the author are the same kind of model using similar prompting approaches, mistakes the author would make will most likely also slip past the reviewer.
In the context of AI-assisted development, "same-origin blind spots" refers to: when the generator and the reviewer share similar model weights, training data distributions and prompt structures, the two highly overlap in their sensitivity to errors.
Human reviewers' blind spots come from limited experience, whereas models' blind spots come from training objectives and data distributions, so swapping in a different person to review is often effective, while swapping in a different prompt often is not.
All Facts (3)
Another camp of commenters worries about "same-origin blind spots": if the reviewer and the author are the same kind of model using similar prompting approaches, mistakes the author would make will most likely also slip past the reviewer.
50%UnverifiedIn the context of AI-assisted development, "same-origin blind spots" refers to: when the generator and the reviewer share similar model weights, training data distributions and prompt structures, the two highly overlap in their sensitivity to errors.
50%UnverifiedHuman reviewers' blind spots come from limited experience, whereas models' blind spots come from training objectives and data distributions, so swapping in a different person to review is often effective, while swapping in a different prompt often is not.
50%