The fundamental problem is that neural nets scale too well. A model trained on a billion dollar GPUs is n times better than a model trained on a $1 million GPU, not to speak of consumer hardware. Unless some efficient form of distributed learning appears, open source will not stand a chance, Except for (niche) applications where smaller brains don't matter. Imagine all these crypto computers would've been used to train models!
The article makes a very simplistic analysis, in that because capitalists won in taking over OpenAI, those concerned about a responsible deployment of very powerful technology lost. But in truth, OpenAI was already a capitalist/commercial enterprise and has been for over a year and those concerned had already failed in keeping the beast contained. The cat is out of the bag, and trying to keep ChatGPT in the box wasn't going to do anything in preventing LLMs to continue to proliferate.<p>Clearly the board should have changed the moment Elon Musk left the company and OpenAI stopped being a non-profit. That's what failed here.<p>If you are going to accept billions of dollars in funding from Microsoft and you are going to start selling a service, then wake up, it's a capitalist venture.<p>The time to pull the trigger, was in releasing ChatGPT to the public. Trying to kick out the CEO after Skynet was activated is pretty pointless.
Much better the capitalists rather than the type of people running the board of OpenAI.<p>This weekend convinced me that these people have no chance of “aligning” even a mildly intelligent AI, much less a superhuman AI.<p>Might as well al least get some useful work out of the AI before it takes over.