From what you wrote, you did it correctly. The way to do this for instanteous velocities is to *estimate* (using that formula, with 2 points that look like they'll give you the right slope) the slope of the tangent line at the point. For the average velocities, you actually plug the two given poits into the slope formula.

There's no "magic" to this other than trying to read the graph as best as you can - there's no other basis than the graph to get the answers.

If the sloppiness of this method - it's lack of precision - makes you uncomfortable (and so it should if you want to be a scientist or engineer someday!), then maybe a better approach (if doing this in "real life", not as an exam question) isn't to use the graph to try to get THE right answer, but rather to use it to BOUND the right answer. Choose points that are "close", but where the "true" slope is obviously greater than when you use those two points. That gives you a lower bound for the "true" slope. Do the reverse to get the upper bound for the slope. Now you have the kind of thing that actually matters to scientists and engineers - not a measured value, but a bounded range that's guaranteed to contain the correct value.