4 ms·
It's crazy how vast and powerful databases of even the most simple things turn out to be. You just store characters and context information when someone types t
by ve55 6y ago
It's crazy how vast and powerful databases of even the most simple things turn out to be. You just store characters and context information when someone types them, and this is the result.
Using this data alongside ML, you could do things such as:
- Identify people based upon their typing style easily, given so much data about them
- Find a lot of secrets that were accidentally pasted into document including passwords, keys, etc
- Try to model the emotions people were feeling as they typed things
- Much, much more, given some data and time
- fouronnes3 6y agoAll valid points, but to be fair you can already do all those things with regular documents.
- bluntfang 6y agosure and i can find out your real time location by hiring a PI to follow you around all day, but it's a lot cheaper and easier if i just pay google for it. plus i cant hire a PI to follow everyone around, there's a cost to that.
- Xenograph 6y agoThe point is that all the data is centralized in a single database in this case.
- BorisMelnik 6y agothe difference being the database in this case is a culmination of god knows how many accounts on the www several TB-PB of data so it bit more scary.
- saadalem 6y agoCarmack2Vec: Identifying and quantifying productivity through keystroke patterns.
- o-__-o 6y agoI studied this vector for attack in college[0] almost 20 years ago based on work done in 1986[1]! [0] - https://dl.acm.org/doi/abs/10.1016/S0167-4048(03)00010-5 https://dl.acm.org/doi/abs/10.1016/S0167-4048(03)00010-5 [1] - https://patents.google.com/patent/US4805222A/en https://patents.google.com/patent/US4805222A/en
- rytill 6y agoI once did an experiment on identifying based solely on keystroke timing: https://github.com/mturnshek/keystroke-timing-identifier https://github.com/mturnshek/keystroke-timing-identifier I did not test it on a large group, but it worked pretty well for a group of ~8 styles and minimal data. I believe a deeper model with enough data could identify on a large scale through timing alone.
- ksaj 6y agoI believe this would work at most scales. Have you noticed how, for example, you can identify a guitar player easily, no matter what guitar they're playing on? When guitar players are first learning or are transcribing, they often slow the music down. If you slow it down enough, you can hear a pattern of the player slowing down and speeding up (behind and ahead of the beat). I believe that has a lot to do with that instant-identifiability, even though it is so hard to detect at speed.