that's true, human are also off-script very often ,the difference is volume and level of off-script, human can generate max 400 LOC a day, that's manageable amount of cognitive load for another human to read comprehend and accept MR, in-out volume match, with 50k LOC , good luck reading that, and from level of off-script, i mean, a human off-script is usually still aligned with original intent, when agent goes off, it goes really off, start installing packages nobody know of or ask for to fix a unit test, that's different, very different
eh, the part about "the only thing between prod and an off-script model is a human reading a diff" -- you realize that humans are also very often off-script, and reading the diff is what we've done for many years as oversight?
that's true, human are also off-script very often ,the difference is volume and level of off-script, human can generate max 400 LOC a day, that's manageable amount of cognitive load for another human to read comprehend and accept MR, in-out volume match, with 50k LOC , good luck reading that, and from level of off-script, i mean, a human off-script is usually still aligned with original intent, when agent goes off, it goes really off, start installing packages nobody know of or ask for to fix a unit test, that's different, very different
eh, the part about "the only thing between prod and an off-script model is a human reading a diff" -- you realize that humans are also very often off-script, and reading the diff is what we've done for many years as oversight?