Specific always beats non-specific, all things being (relatively) equal, i.e a world class cyclist would always perform better at cycling than a world class runner and vice versa. However a world class endurance athlete (in any sport) would likely outperform a recreational athlete (trained at only a low level) at their sport due to overall conditioning etc etc
For the most part, I agree, but then it comes down to what is truly "specific".
I think specificity exists on a bit of a continuum. Here's an example that I think gets across what I am trying to say.
In rowing, the 2k is a pretty standard distance; let's use 7:00 as the target time to establish the time domain of the event.
Specificity in that case is rowing at 2k target pace (1:45/500m, or 302w, give or take) with target stroke rate, let's just establish r30 as the target for example. Anything besides 302w for 7:00 at r30 is not actually specific. The modality is specific but there are other things that are not - pace/power, stroke rate, time domain, etc.
I don't think anyone would make the argument that the only way to improve a 2k is by doing 2k's - that is one aspect of training but not the only thing you need to do.
You could train at a similar stroke rate (r30) but different target pace, you could train at a similar pace (302w) but with a different stoke rate, or you could train at a similar SPI (watts per stroke; 10.x); all are using rowing as a specific modality but they are one step removed from a 7:00 2k at r30.
Now going one step further away, you can train with various paces, stroke rates, or SPI's. Again, all using rowing as the modality but not necessarily specific to the actual 7:00 2k at r30.
All of this might help to build a 7:00 2k at r30 but it's not necessarily specific.
There are some who make the argument that by training with a specific modality but in a way that is non-specific to the target, you can experience a bit of technique spoliation for a lack of better words. Klaas Lok speaks about this in a way in "The Easy Interval" method where you need to run fast to run fast; you can't just plod along without regularly touching on the specific target. Nils van der Poel notes something similar in "How to Skate a 10k". They take the position that, if you are going to move away from specificity to train aspects of what is needed for the event, then it's better to do that with different modalities to preserve technique.
Using the 7:00 2k at r30; you need a certain amount of several physical attributes - aerobic capacity, strength, power expression, etc.
There is definitely something to be said about using the modality to get better at the modality.
However, in order to develop the aerobic capacity needed for a 7:00 2k at r30, training rowing at lower pace / power, lower stroke rate, or lower SPI might spoil the exact technique needed for the target so it might be better to do the aerobic training with a different modality like cycling, running, etc. In order to develop the power needed, it might be better to focus on a different modality like actual strength training in the weight room as opposed to rowing at a higher drag factor (ie. rowing with more resistance) because rowing with a different drag factor can impact the speed of the stroke and thus the technique, stroke rate, etc.
I'm not saying I agree 100% with the Klaas Lok and Nils van der Poel stuff through and through but I do think there is some merit. Do you need to row to improve rowing? 100% yes. After a certain amount of rowing though, is it better to row more, or is it better to do other things that improve the qualities needed for a certain rowing performance?