[Max RNR]

Give this a shot:

https://cloud.google.com/vision/ **

Copy some pics into the provided form and see what returns. I'd suggest to start with obvious pics and then get more and more obscure. Also, I recommend that you do this while your Google account is not logged in and other security measures (vpn, vm, etc.) are in place.

I was impressed.

** This may be old news, but a forum search did not provide any hits on "google vision" or "google vision api" (both in quotes), and word on the street (of Nerd City, south of 49th Ave) is that YouTube uses this AI when processing/flagging video uploads.

[Mage]

Don't feed the beast any images!

[Max RNR]

(12-16-2018 05:49 PM)Mage Wrote:  

Don't feed the beast any images!

There are two basic ways to be invisible. One is to not be seen by stealth. The other is to understand how your enemy sees and to exploit it's weaknesses so as not to be seen.

The argument can be made that by training against the beast you will become the beast, however, so will it become you.... and therefore your advantage can be made clear by using a never before seen vision when unexpected.

[Laner]

I put in a photo of a Mexican chick I banged. I couldn't remember her name but in less than a second I had her full name, where she was from, and a bunch of other pics of her as well as places to get her social media accounts.

I need a drink.

[Caduceus]

Looks like you need to be logged into google services to use this.
No thanks, I'm not that gullible.

[Max RNR]

And then the girls at Ars Technica chimed in against the boys (and duly rewrote the article -- taking out all the AI graphs) a day later:

https://arstechnica.com/science/2018/12/...-images/4/

"But beyond that, convolutional networks have no real understanding of geometry. They can't recognize that one image is similar to another if it's rotated by 45 degrees or scaled up by a factor of two. Convolutional networks make no attempt to understand three-dimensional object structures, and they can't correct for varying lighting conditions."

Natch.

"Open the pod bay doors, Hal."

[Horus]

When I clicked login to see what would happen, it took me to a login screen from my old university which I haven't accessed for years. Seems extremely dodgy.

[Winston Wolfe]

(12-17-2018 12:58 PM)Laner Wrote:  

I put in a photo of a Mexican chick I banged. I couldn't remember her name but in less than a second I had her full name, where she was from, and a bunch of other pics of her as well as places to get her social media accounts.

I need a drink.

How does this work?

I tried it out too by uploading a photo of a chick. I am the only one who has this photo and it has not been shared with anyone in the past.

The tool does a pretty good job identifying properties of her face and the surroundings and even the most likely emotional state of the girl. However it does not give me any names or other pics (I know she does have an internet presence, when I google her name I find various photos and social media accounts).

[Max RNR]

(12-20-2018 06:38 PM)Max RNR Wrote:  

And then the girls at Ars Technica chimed in against the boys (and duly rewrote the article -- taking out all the AI graphs) a day later:

https://arstechnica.com/science/2018/12/...-images/4/

"But beyond that, convolutional networks have no real understanding of geometry. They can't recognize that one image is similar to another if it's rotated by 45 degrees or scaled up by a factor of two. Convolutional networks make no attempt to understand three-dimensional object structures, and they can't correct for varying lighting conditions."

Natch.

"Open the pod bay doors, Hal."

Here is the original.

https://arstechnica.com/science/2018/12/...ng-images/

Good and thought provoking reading if you are of intermediate math or science mind.