The big wow will come when Metadata, an which is for now fragmented not only in within various training models , but also within a model or entity ( researcher perspective, engineering is of a potential barrier, and users ) is happening.
You get some unexpected results using a dataset in a different ML model .
Unifying that data ( oversimplified as "tagging) will facilitate an explosion.
Right now is VHS, vs betaMax vs Umatic - that lead to DVD, Hd format ( remember sony's beta revenge with PS Blueray, 4k, 8k 8k/120 fps. )
We're still at the very early fragmentation.
A factor for meaningful inference is using a combination of Machine Learning and Human verification in production of that. The Human Verification part is in poor modelling also left to a self feeding MLOps.
That's why you see sometimes hilarious outcomes.
As you and everybody playing with that noted, in order to get to performance values which can actually be purposefully meaningful. Hence , that will be a real value to be sold . ( I predict the next inflatable buzz balloon will play on that)
Right now we are still using separate database pipeline that we patch to create a larger access to tagged datasets.
Which , as of now, is painfully difficult ( costly) difficult or just impossible to query across different databases, ( incongruous architecture, formating, language, regions)
One solution proposed ( there's more) but this seems decent is a Graph Database.
One exemple is RocksDB
Read for more here if you are interested.
Trivia:
Deep Learning - was proposed in Montréal's MILA institute by Joshua Bengio. Improved significantly the ML.
Just , as a need for what I was doing got a basic certification - then I realized that's waaay above my head mathematically and very quickly misunderstood and misused for false inferences ( GIGO rule).
Watching from the side. Bengio is trying to point to that, but he's over shouted by the hedge funds and few in the real ML , bayesian circles get his point.
That's where the beauty stands. And the inquisitive drive that leads to creativity.
There's disadvantages to uniformity. ( see the results of ML training on the saaame available and rigidly tagged ( simplified filler word for metadata) databases.
I had my battles with AI back in 2016, especially after seeing "Sophie the humanoid robot" live in one of the many tech 4.0-related events I attended, discussing about the future of work and the pros/cons of all these new tech.
At that time, I was sure that AI is a tool, like you mentioned, a very powerful tool that can make or break the next generation of humanity.
And like any powerful tool comes great responsibilities.
I came to conclude then, there is no escape to how AI will be part of our lives, just like electricity, computers and airplanes.
How useful AI really therefore, truly depends on how we human beings use them and thus the debate or discussion of ethics have to come in. As designers, creators and artists use AI, I feel that we really have to think wider beyond just mindless consumption, but ask deeper and more intentional questions such as, " How can I design the use of AI to be collaborative and not dependent on it?", "Will what I design be sustainable for the systems in place?"
I am not sure if we will have unnecessary problems or solutions with the increasing use of AI. Only time will tell.
One thing I can say for certain is, AI will amplifies the cracks of society, as well as the greatness of humanity. We truly are at the edge of a new era.
P.S. I am still contemplating if I want to use AI to help with writing and art-creation too.
Happy Holidays J.J.
Hope 2022 have been awesome to you in Taiwan if you are there.
Happy holidays to you too Yan! As for your contemplation, I'd say give it a shot and treat is as a tool not a replacement. The latter mindsight I think is the only way to get truly adept at using it.
The big wow will come when Metadata, an which is for now fragmented not only in within various training models , but also within a model or entity ( researcher perspective, engineering is of a potential barrier, and users ) is happening.
You get some unexpected results using a dataset in a different ML model .
Unifying that data ( oversimplified as "tagging) will facilitate an explosion.
Right now is VHS, vs betaMax vs Umatic - that lead to DVD, Hd format ( remember sony's beta revenge with PS Blueray, 4k, 8k 8k/120 fps. )
We're still at the very early fragmentation.
A factor for meaningful inference is using a combination of Machine Learning and Human verification in production of that. The Human Verification part is in poor modelling also left to a self feeding MLOps.
That's why you see sometimes hilarious outcomes.
As you and everybody playing with that noted, in order to get to performance values which can actually be purposefully meaningful. Hence , that will be a real value to be sold . ( I predict the next inflatable buzz balloon will play on that)
Right now we are still using separate database pipeline that we patch to create a larger access to tagged datasets.
Which , as of now, is painfully difficult ( costly) difficult or just impossible to query across different databases, ( incongruous architecture, formating, language, regions)
One solution proposed ( there's more) but this seems decent is a Graph Database.
One exemple is RocksDB
Read for more here if you are interested.
Trivia:
Deep Learning - was proposed in Montréal's MILA institute by Joshua Bengio. Improved significantly the ML.
Just , as a need for what I was doing got a basic certification - then I realized that's waaay above my head mathematically and very quickly misunderstood and misused for false inferences ( GIGO rule).
Watching from the side. Bengio is trying to point to that, but he's over shouted by the hedge funds and few in the real ML , bayesian circles get his point.
I get the feeling we each know very different things in life haha
That's where the beauty stands. And the inquisitive drive that leads to creativity.
There's disadvantages to uniformity. ( see the results of ML training on the saaame available and rigidly tagged ( simplified filler word for metadata) databases.
I had my battles with AI back in 2016, especially after seeing "Sophie the humanoid robot" live in one of the many tech 4.0-related events I attended, discussing about the future of work and the pros/cons of all these new tech.
At that time, I was sure that AI is a tool, like you mentioned, a very powerful tool that can make or break the next generation of humanity.
And like any powerful tool comes great responsibilities.
I came to conclude then, there is no escape to how AI will be part of our lives, just like electricity, computers and airplanes.
How useful AI really therefore, truly depends on how we human beings use them and thus the debate or discussion of ethics have to come in. As designers, creators and artists use AI, I feel that we really have to think wider beyond just mindless consumption, but ask deeper and more intentional questions such as, " How can I design the use of AI to be collaborative and not dependent on it?", "Will what I design be sustainable for the systems in place?"
I am not sure if we will have unnecessary problems or solutions with the increasing use of AI. Only time will tell.
One thing I can say for certain is, AI will amplifies the cracks of society, as well as the greatness of humanity. We truly are at the edge of a new era.
P.S. I am still contemplating if I want to use AI to help with writing and art-creation too.
Happy Holidays J.J.
Hope 2022 have been awesome to you in Taiwan if you are there.
Happy holidays to you too Yan! As for your contemplation, I'd say give it a shot and treat is as a tool not a replacement. The latter mindsight I think is the only way to get truly adept at using it.
Hey there. JC. Where'd you get that cool image? Thanks!
The main one? I used Midjourney for that