
One weight moves the two inputs in opposite directions, so it cannot shift both positively. This exact counterexample assumes a bias-free linear model; an editable bias changes what is possible.
Settings, scope and source files
Exact counterexample at . A shared weight update yields , so the direct request is unreachable for . An editable bias would change this conclusion.
Plays once when in view · pause at any time