Wow, awesome presentation. Just want to comment on Alpha Zero difference versus AlphaGo: AlphaGo has a memory, a database stored but when you let that database to be cloned and then let it improve through learning, and again introduce a competition from which you choose a winner, this is essentially a genetic algorithm. All life including us humans is a combination of neural networks and genetic algorithms. If the learning experience is coded in a certain storage device let's call it a gene and then you clone it or even more you introduce sex which means you take two genes and combine them and you let them improve through random experiences, then you introduce a competition and select only the winners, this is the fastest way to iterate learning towards human level intelligence. I recommend you reading The Red Queen by Matt Ridley (can be found the pdf on Google), he talks about that in chapter 4, this book really blew my mind.
This is an extremely interesting and insightful video. Really great breakdown and analysis of the current state of things. Thanks and kudos 👍 Also, you sound a lot like Matthew Rhys.
There is still an existential difference between brain function (whether synthetic or otherwise) responsible for performance and that, responsible for mind which, to date, remains unaddressed. Failure to appreciate this distinction will introduce needless consequences.
its been less than 2 years since this video was uploaded, yet the data used seems antiquated already. Dactyl, gpt-2, tesla autopilot, deepfakes... the field is advancing hyperexponentially.
Wow, awesome presentation. Just want to comment on Alpha Zero difference versus AlphaGo: AlphaGo has a memory, a database stored but when you let that database to be cloned and then let it improve through learning, and again introduce a competition from which you choose a winner, this is essentially a genetic algorithm. All life including us humans is a combination of neural networks and genetic algorithms. If the learning experience is coded in a certain storage device let's call it a gene and then you clone it or even more you introduce sex which means you take two genes and combine them and you let them improve through random experiences, then you introduce a competition and select only the winners, this is the fastest way to iterate learning towards human level intelligence. I recommend you reading The Red Queen by Matt Ridley (can be found the pdf on Google), he talks about that in chapter 4, this book really blew my mind.
Bravo, wonderful job on this work. Those are some good insights to ponder on.
Thanks for upload, i really like both the style and topics of your presentations.
Please consider making a version of this graph with dots that appear with the progress of time in the last 100 years
This is an extremely interesting and insightful video. Really great breakdown and analysis of the current state of things. Thanks and kudos 👍
Also, you sound a lot like Matthew Rhys.
like will be given.
Great video and narration! Just a quick correction -- Deep Blue beat Kasparov in 1997.
There is still an existential difference between brain function (whether synthetic or otherwise) responsible for performance and that, responsible for mind which, to date, remains unaddressed. Failure to appreciate this distinction will introduce needless consequences.
Where would you place IBM's Watson on the graph?
Were do you see evolutionary algorithms displayed on this graph?
There is an error. The const volume curve on a log-scale graph should be a straight line.
I would describe it as human ability, not human intelligence. Alpha zero will not choose to learn something new to adapt or be deprecated.
its been less than 2 years since this video was uploaded, yet the data used seems antiquated already. Dactyl, gpt-2, tesla autopilot, deepfakes... the field is advancing hyperexponentially.
26:20
Haha. Weeeb.