Oxford scientists develop extremely rapid diagnostic test for Covid-19

Could someone explain me this. The authors say (in the preprint [1]):

"The trained network was able to differentiate positive and negative IBV samples with high accuracy (91.4%), sensitivity (91.9%) and specificity (90.9%) (Fig.2G). Of note, these probabilities refer to single virus particles in the sample and not the whole sample; the probability of correctly identifying a sample with hundreds or thousands of IBV virus particles will therefore approach 100%. "

But then when the actual lab tests are performed:

"The trained network was able to distinguish between virus-positive and virus-negative samples with excellent accuracy, distinguishing between SARS-CoV-2-positive and negative BBXs with an accuracy of ~70% (Fig.3A)".

An accuracy of 70% is totally useless for such tests if widely applied, right? What am I missing here?

[1] https://www.medrxiv.org/content/10.1101/2020.10.13.20212035v...

Link isn’t working, did a quick search for alternative : https://www.msn.com/en-us/health/medical/oxford-university-s...

This is pretty much how high throughput gene sequencing works, right?