True, it's a union of all these different fields coming together, might not be the best AI toy out there, but it's definitely something we can buy for a reasonable price. Imagine the kids growing up with such toys, they know how machines learn, it becomes part of their growing up process..
That is a very interesting idea, how kids interact with AI toys in the long term. I remember when I was little parents alway complain how children would abandon their new toys after a few days. Making a new generation of toy that keep kids interested with AI seems really important for toy companies.
Is there meant to be something on the second 'page' (ugh I hate this single-page-permascroll-with-fixed-header layout that they all do these days) other than the 'let there be more light' box? It's just a grey background so I suspect something didn't load there. Otherwise I just see a generic video showing object tagging on the first 'page' and the pretty-but-pointless graph thing on the third 'page'.
Firefox 59.0.2 with PrivacyBadger and uBlock Origin (although I turned both off and still nothing loaded).
There's an object detection library (YOLO[1]) freely available that achieves similar bounding boxes/matches to ambient's homepage demo; Joseph Redmon also gave a TedX[2] about it that describes some of the technical details.
I thought augmented reality with mobile camera was going to take off after seeing that demo. But 8 years later, all we have are snapchat filters (and pokemon go for a brief few weeks).
Hopefully ARKit/ARCore will make it easier to create the next the-future-is-now level of awe-inspiration app.
Pokemon Go still has millions of active daily players, more than many games that have e-sports tournaments. It's in active development. New features are added all the time. Sorry it's not for you, but pretending that it was a flash in the pan is just ignorant.
>The output was put to work in what Mr Nix called "behavioural micro-targeting" and "psychographic messaging".
Instead of creating broad messages to talk to the voters in bulk - they could create micro-campaigns that could target particular voters and push their buttons (and that might have raised flags done publicly). Sometime this meant influencing someone borderline towards Trump. Sometimes this meant targeting Clinton supporters in swing-state to influence them not to vote (https://www.theverge.com/2016/10/27/13434246/donald-trump-ta...)
This adds up to a far more weaponized media than traditional advertising could be.
That same AI drove me away. I subscribe to 3 hobby-related groups and because virtually all of my family and friends no longer post anything on Facebook, I only ever see content related to that hobby.
If that description is honest, why do you blame the algorithm for selecting hobby-related content from a set of inputs that only contains hobby-related content? That's like buying a football magazine, then complaining that it's only football, football, football in there.
AI has been a moving goalpost since its conception, all the way from
• "it can play tic tac toe";
• to "it can beat people at chess, Jeopardy, Go, DotA";
• to "it can see, understand and physically move around our world";
• to "it solves problems the smartest humans can't begin to fathom"
Spotify's machine learning algorithms are definitely beyond tic tac toe as they can learn from millions of listeners and look at a single listening history to figure out great music suggestions. I would put that under the AI umbrella as it's something people were doing before computers did it.
What is "smartly comparing lists"? To describe further (as I'm an ex-Spotify employee I have some outdated insight into how it works):
Spotify has historically used machine learning to tweak a predictive engine that can convert a track or artist into an N-dimensional value and then use the distance to other tracks/artists in this N-dimensional space.
Is that AI? Maybe. How does the brain work? Maybe when you see a dog it's converted into an N-dimensional space where cats are pretty close, at least much closer than turtles. So if that is human intelligence, is Spotify's recommendation engine not artificial intelligence?
I'm working on project that uses distances in N-dimensional space for determining appropriate human response and face expression; I was also wandering if this falls under AI umbrella.
Spotify will tell artist to change or remix parts of their songs. Songwriting is a craft but on a platform like spotify its turned into a science. Think a/b testing for songs. Thats what i have heard they are doing.
It’s a machine (algorithm) that looks and learns from my listening history and that of other users, defines certain specific listening patterns/genres for me and it compiles mixes with old and new music for each of those genres.
I don’t see that much difference with an AI that looks at and learns from a lot of historic go games to recommend or play the optimal next moves for the current go board.
This might seem like a flippant comment but for me working in a computer museum watching the way visitors interacted with eliza was fascinating. theres something about how well-crafted responses can charm if used in the right way that some more complicated AI such as alexa seem to lack so far.
Shameless plug, I used it in a human tracking video camera. https://github.com/GistNoesis/Linn-Photobooth
It's actually kind of scary once you realize how hard it is to escape being tracked. It's even more scary once you realize that it can be done by any geek as a week-end project. Just mod it with an AR15, and you have your own Portal turret.
Edit: I misread densepose for deeppose which my project use. Will give densepose a try.
There was a project [1] on synthesizing photo-realistic textures from labels of an image. There are actually a lot of these projects (Pix2Pix [2] is another famous one) but the thing that made it memorable is one of their demo application.
They took a video of GTA (a car driving game) and reskinned to be photo-realistic and I couldn't even tell that it was an ingame footage instead of a real car driving through a city: https://youtu.be/0fhUJT21-bs?t=124
They used to be popular like 10+ years ago, so I don't remember the URL, There's a test called "turing test". I've a vague memory that the bot I tested was called Alice. I tried Google but got flooded with crap. It seems the market has been monetized, the bots I tested 10 years ago where much more impressive then what I can find now. They learned by talking to real humans so it's possible that the public ones have been abused in order to get them to say nasty things.
The fastest and most accurate way to track time for teams. Automatic time tracking helps you never forget what you worked on. Smart resource scheduling assures everyone's time is utilized. Powerful reporting helps you identify opportunities to improve productivity.
The product is "Timely"
follow on
https://www.besanttechnologies.com/training-courses/data-war...
71 comments
[ 3.9 ms ] story [ 194 ms ] threadFirefox 59.0.2 with PrivacyBadger and uBlock Origin (although I turned both off and still nothing loaded).
[1]: https://pjreddie.com/darknet/yolo/
[2]: https://www.ted.com/talks/joseph_redmon_how_a_computer_learn...
Had google glass not stored video and processed everything in real-time it could have avoided the "glasshole" stigma.
People would have likely mocked it for not being able to capture video, something they could have added in later in "response to demand".
The world doesn't need more omni-present recording devices, but solid AR would be a net benefit.
Hopefully ARKit/ARCore will make it easier to create the next the-future-is-now level of awe-inspiration app.
The teachable machine from Google is a great little experiment.
https://google.github.io/tacotron/publications/tacotron2/ind...
Blog post: https://research.googleblog.com/2017/12/tacotron-2-generatin... Paper: https://arxiv.org/abs/1712.05884
Check it out.
Take that as you please.
Advertising in tv is fine but on fb it is gaming the system??
>The output was put to work in what Mr Nix called "behavioural micro-targeting" and "psychographic messaging".
Instead of creating broad messages to talk to the voters in bulk - they could create micro-campaigns that could target particular voters and push their buttons (and that might have raised flags done publicly). Sometime this meant influencing someone borderline towards Trump. Sometimes this meant targeting Clinton supporters in swing-state to influence them not to vote (https://www.theverge.com/2016/10/27/13434246/donald-trump-ta...)
This adds up to a far more weaponized media than traditional advertising could be.
The amount of "news" coverage, the amount of money invested, television, radio, with much more reach. The strong opinions.
It is like brexit some people don't want to believe it.
Face and object recognition helped me quite a lot in finding old photos.
In all honesty, I think the lines are severely blurred at this point.
• "it can play tic tac toe";
• to "it can beat people at chess, Jeopardy, Go, DotA";
• to "it can see, understand and physically move around our world";
• to "it solves problems the smartest humans can't begin to fathom"
Spotify's machine learning algorithms are definitely beyond tic tac toe as they can learn from millions of listeners and look at a single listening history to figure out great music suggestions. I would put that under the AI umbrella as it's something people were doing before computers did it.
Spotify has historically used machine learning to tweak a predictive engine that can convert a track or artist into an N-dimensional value and then use the distance to other tracks/artists in this N-dimensional space.
Is that AI? Maybe. How does the brain work? Maybe when you see a dog it's converted into an N-dimensional space where cats are pretty close, at least much closer than turtles. So if that is human intelligence, is Spotify's recommendation engine not artificial intelligence?
I'm working on project that uses distances in N-dimensional space for determining appropriate human response and face expression; I was also wandering if this falls under AI umbrella.
https://www.google.com/amp/s/amp.fastcompany.com/40439000/wh...
I don’t see that much difference with an AI that looks at and learns from a lot of historic go games to recommend or play the optimal next moves for the current go board.
https://www.youtube.com/watch?v=RuAp92wW9bg
Edit: I misread densepose for deeppose which my project use. Will give densepose a try.
Nice background videos showing some demonstrations of video analysis and detection for safety or other application.
They took a video of GTA (a car driving game) and reskinned to be photo-realistic and I couldn't even tell that it was an ingame footage instead of a real car driving through a city: https://youtu.be/0fhUJT21-bs?t=124
[1] http://cqf.io/ImageSynthesis/ or https://www.youtube.com/watch?v=zfKrtQur3Lw [2] https://affinelayer.com/pixsrv/
https://getmagic.com/
It's truly universal; its applications encompass most of the other examples I've seen posted here.