We simulate lattice QCD at finite quark-number chemical potential, $\mu$, using the complex-Langevin equation (CLE) with gauge-cooling and adaptive updating to prevent instabilities. The CLE is used because QCD at finite $\mu$ has a complex fermion determinant which precludes the use of standard simulation methods based on importance sampling. Since, even when CLE simulations converge, they are not guaranteed to produce correct results except under very stringent conditions, which lattice QCD at finite $\mu$ does not obey, we need extensive testing to determine under what conditions it produces reliable results. We performed simulations at $\beta=6/g^2=5.6$ and $\beta=5.7$, both at $m=0.025$. For small $\mu$ and $\mu$ large enough to produce saturation, measured observables appear to be approaching their correct values as the coupling is decreased. However, for intermediate $\mu$ values, these simulations predict a transition from hadronic to nuclear matter at a $\mu$ which is far too small. Since there is evidence that for CLE simulations to produce correct results the trajectories should remain close to the $SU(3)$ manifold (at least for small $\mu$), we explore the parameter space to see where this is true. We find that the distance from this manifold decreases as the coupling decreases and as the quark mass (in lattice units) decreases, i.e. as we approach the continuum limit. This indicates that we need to simulate at smaller couplings and quark masses (requiring larger lattices) to see if these can produce the correct physics.