We measure the angular two-point correlation function of sub-millimeter galaxies (SMGs) from 1.1-millimeter imaging of the COSMOS field with the AzTEC camera and ASTE 10-meter telescope. These data yields one of the largest contiguous samples of SMGs to date, covering an area of 0.72 degrees^2 down to a 1.26 mJy/beam (1-sigma) limit, including 189 (328) sources with S/N greater than 3.5 (3). We can only set upper limits to the correlation length r_0, modeling the correlation function as a power-law with pre-assigned slope. Assuming existing redshift distributions, we derive 68.3% confidence level upper limits of r_0 < 6-8 h^-1 Mpc at 3.7 mJy, and r_0 < 11-12 h^-1 Mpc at 4.2 mJy. Although consistent with most previous estimates, these upper limits imply that the real r_0 is likely smaller. This casts doubts on the robustness of claims that SMGs are characterized by significantly stronger spatial clustering, (and thus larger mass), than differently selected galaxies at high-redshift. Using Monte Carlo simulations we show that even strongly clustered distributions of galaxies can appear unclustered when sampled with limited sensitivity and coarse angular resolution common to current sub-millimeter surveys. The simulations, however, also show that unclustered distributions can appear strongly clustered under these circumstances. From the simulations, we predict that at our survey depth, a mapped area of two degrees^2 is needed to reconstruct the correlation function, assuming smaller beam sizes of future surveys (e.g. the Large Millimeter Telescope's 6" beam size). At present, robust measures of the clustering strength of bright SMGs appear to be below the reach of most observations.