Hacker News new | past | comments | ask | show | jobs | submit login

Is this AI like detecting QR is an AI task or AI could actually mislabel product some percentage of time?



I recall the article mentioned they got to "99% reliability". As a manufacturer of high-end products looking for a good authentication solution, I was thinking this might be cool until I read that. Having errors in 1 in 100 identification passes is nowhere near usable. I'd want I'd down under one misidentification in 10k or 100k readings (fail/redo/succeed is ok in 1%). The entire point of the system is to be able to definitely determine that "this is a counterfeit" or "yes, we made this in batch K69342". If it can't do that reliably, why would I invest in building this into our manufacturing flow?


Reading the content it seems like 99% is only for detection of physical tampering, i.e taking the chip out of real thing and putting it in fake thing, which significantly decreases the intent tamper as you need to destroy the real thing.


Wow, if it's only 99% catching that condition, it seems even worse! (I read it as 99% re-recognizing un-mangled items, but yes, if it is 99999/100k re-recognizing a good tag, it might be ok to let pass 1/100 counterfeits).




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: