the.com/steering
The knob that quietly decides whether an AI is helpful or horrifying.
means The act of guiding the direction, movement, or course of something — literally turning a vehicle's wheel, or figuratively nudging a system, conversation, or organization toward an intended outcome.
from Traces back to Old English 'steoran,' meaning to steer or guide, rooted in the Proto-Germanic 'stiurjan.' The deeper root ties to a word for a stiff or rigid support — the same family that gives 'stern,' as in the back of a ship, where the steering apparatus traditionally lived. So the word carries an old nautical soul: the hand on the rudder, deciding whether the ship meets the harbor or the rocks.
- no retraining
- works by editing activations, not weights
- golden gate claude
- anthropic made a model obsessed with one bridge
- vector arithmetic
- concepts behave like directions you can add
- double edged
- same trick can suppress refusals or safety