CIE-LAB, or anything that is based on how humans see, rather than the phosphors in old CRTs. HSV/L are just a simple linear transformation of RGB, which corresponds to the phosphors available to produce CRTs, not the vision properties of the human eye. HSV/L was a good approximation when computers were too slow to do the transformations, but it's an appallingly bad space for doing automated color palette generation.
HSV is ok for color pickers where a human is picking a color, but it makes no sense when you are picking the H angle and generating color based on it, because a constant H angle delta does not correspond to a constant change in human perception at different base angles. In other words, if you have a base color at some X degress of H, and two accent colors at X+30 and X-30, you won't perceive the two accents as being the same distance from the base, X, because different H angles are perceived as different color distances at different points in the spectrum. The transformation from H to perception space is not linear.
And that's just hue. L, S, and V (pick two!) have their own issues, with different hues being perceived as different lightness or saturation at the same L or S value.
HUSL was already mentioned, so I bring up HSP [1]. The idea is to adjust the lightness (the L in HSL) by something which is closer to the percieved luminescence.
Also I wonder if HSL will ever manage to replace HSB/HSV as the default choice for color pickers.
