223
OpenAI says it’s “impossible” to create useful AI models without copyrighted material
(arstechnica.com)
A nice place to discuss rumors, happenings, innovations, and challenges in the technology sphere. We also welcome discussions on the intersections of technology and society. If it’s technological news or discussion of technology, it probably belongs here.
Remember the overriding ethos on Beehaw: Be(e) Nice. Each user you encounter here is a person, and should be treated with kindness (even if they’re wrong, or use a Linux distro you don’t like). Personal attacks will not be tolerated.
Subcommunities on Beehaw:
This community's icon was made by Aaron Schneider, under the CC-BY-NC-SA 4.0 license.
Or, or, or, hear me out:
Maybe their particular approach to making an AI is flawed.
Its like people do not know that there are many different kinds of ways that attempt to do AI.
Many of them do not rely on basically a training set that is the cumulative sum of all human generated content of every imaginable kind.
What ways do you mean? More than just expert-systems, I'd imagine.
Well, off the top of my head:
Whole Brain Emulation, attempting to model a human brain as physically accurately as possible inside a computer.
Genetic Iteration (not the correct term for it but it escapes me at the moment), where you set up a simulated environment for digital actors, then simulate quasi-neurons, quasi-body parts dictated by quasi-dna, in a way that mimics actual biological natural selection and evolution, and then you run the simulation millions of times until your digital creature develops a stable survival strategy.
Similar approaches to this have been used to do things like teach an AI humanoid how to develop its own winning martial arts style via many many iterations, starting from not even being able to stand up, much less do anything to an opponent.
Both of these approaches obviously have drawbacks and strengths, and could possibly be successful at far more than what they have achieved to date, or maybe not, due to known or existing problems, but neither of them rely on a training set of essentially the entirety of all content on the internet.