I just realize that in the tutorial example of "geometric accuracy", recon-all -autorecon2-wm is used to recreate the final surfaces; while in "lateral ventricle", recon-all -autorecon2-wm -autorecon3 is used. However, the modification steps look the same to me -- fill in the "hole" of white matter in wm.mgz. 

Why do they use different commands after all? What's the difference between -autorecon2-wm and -autorecon2-wm -autorecon3?

Thank you very much!

Sincerely,

Ye