def compute_loss(predictions, targets):
    """Mean squared error loss"""
    return np.mean((predictions - targets) ** 2)  # Square the difference to penalize large errors