25 ms·

I recently encountered the same underlying problem with Google's knowledge graph. I do a lot of scientific image analysis using an ancient (but reliable!) piec

I recently encountered the same underlying problem with Google's knowledge graph. I do a lot of scientific image analysis using an ancient (but reliable!) piece of software called ImageJ [0]. There's a more recent distro of the same called FIJI [1]. So when I tried looking for how to extract EXIF data for GPS coordinates using ImageJ (not even mentioning FIJI), Google returned an info box about the Fiji-the-nation and provided the coordinates of said nation: https://i.imgur.com/HxSh8Zv.png https://i.imgur.com/HxSh8Zv.png [0]: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5554542/ https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5554542/ [1]: https://imagej.net/software/fiji/ https://imagej.net/software/fiji/

tgbugs 5y ago
Glorious. False positives on the synonym match gone wild. This is the kind of thing that deep learning was supposed to solve. The fact that it is failing from specific (imagej) to general (fiji) is beyond the wildest dreams of anyone doing quality assurance. I'm guessing the neural nets have been slacking by trying to do pure substitution matches and the devs have been rewarding them at the worst possible time. Wow. This is like seeing a toddler who is screaming in public and the parents are actively giving them candy to try and get them to stop. Wow. In case anyone doubts that this is real I'm seeing it too https://i.imgur.com/PXepl6C.png https://i.imgur.com/PXepl6C.png
- eitland 5y ago
  This started going downhill about a decade ago. I wrote about it 8 years ago and I am fairly sure I had seen it going on for a while already at that time: https://techinorg.blogspot.com/2013/03/?m=0 https://techinorg.blogspot.com/2013/03/?m=0