Here is a short explanation:

Feedbacks in the form of one model feeding back into its sub-model's runs, potentially reinforcing, if not outright enhancing, existing errors in judgment.

GFS & GFS-based runs are also creating arguably over-emphasized convective feedback in the form of auxiliary MCSs over Texas and over the northeastern GOM/southeastern states ... these often tend to diminish the natural state of things, throwing model forecasts off in terms of intensity, direction, timing, and rainfall locations and amounts.

Modeled building in of a blocking subtropical high looks somewhat more influential than what is actually happening, so far.

Alex is a deeper system than models have generally been capturing, and is being directed to a much greater degree by steering currents that would take the system farther north, before having the opportunity to begin shoving Alex to the left (let alone southwest), than by steering currents that would not have as much of a northern bent, in a less vertical, less deep system.