Let's try to understand what is being asked. First let's write the problem in matrix notation:
[eqn]
\left[ \begin{array}{ccc} 1 & 3 & 4 \\ 2 & 1 & 3 \\ -1 & 2 & 1 \\ \end{array} \right]
\left[ \begin{array}{c} x \\ y \\ z \\ \end{array} \right]
=
\left[ \begin{array}{c} 0 \\ 0 \\ 0 \\ \end{array} \right]
[/eqn]
You are tasked to find the solution set for this system of equations. That means you want to find every [x;y;z] (column) vector that satisfies the equation.
It is a basic fact of LA that the solution set to a linear system either has 0 solutions, 1 solution, or infinite solutions. If you did not know this, we can go over the proof.
So the question is asking you to find a basis for this solution set of (column) vectors, so necessarily the basis will be composed of column vectors. Capiche?
To solve systems of equations, we often use elimination, as you attempted. Unfortunately you didn't correctly calculate the Reduced Row Echelon form of the matrix (don't sweat it, it's really easy to make an arithmetic mistake). The matrix you should arrive at is:
[eqn]
\left[ \begin{array}{ccc} 1 & 3 & 4 \\ 0 & 0 & 0 \\ 0 & 0 & 0 \\ \end{array}\right]
[/eqn]
Now that the matrix is in RRE form, it should be immediately obvious that there is only one free parameter (if not, review elimination and free parameters). That tells us that the dimension of the solution set is 1, or in other words a line. To calculate a basis for this line, we just have to find one non-zero vector on the line; any other solution is simply a scaled copy of the basis vector.
For arbitrary problems, you finish the full blown Gauss-Jordan Elimination algorithm to get an answer. In this case, it's not hard to see that only the first equation matters from the RRE form, and it's easy to see by inspection that one solution to this system is [-1;-1;1]. Granted, any non-zero scalar multiple of this vector is also a valid basis vector, so something like [2;2;-2] is also a valid answer.
Make sense?