the double voice: teaching ai’s to doubt themselves
how to use llm’s that know what they don’t know
every senior engineer knows the moment.
you’re staring at cursed legacy code.
not because you found the bug but because you found something worse.
it works.
& you don’t know why.
“this shouldn’t work,” you mutter, watching a function violate everything you believe about good code while passing every test.
& in that moment, you’re holding two truths: what the code does & what you don’t understand about why it does it.
that’s intelligence.
not answers, awareness of unknowing.
most ai systems today are like junior devs who don’t tell you explicitly about what they doubt.
they’re fluent, fast, confident.
& seemingly unaware of what they don’t know.
the two voice technique
