I mean that's a pretty big endorsement and something people here should think about. Also I read the highlights of the paper linked here and it seems fine and

I mean that's a pretty big endorsement and something people here should think about. Also I read the highlights of the paper linked here and it seems fine and reasonable. These are reasonable questions to ask. Sure we can mitigate the cost of training these models with green energy or finding new optimization./compression techniques etc... but that doesn't mean that with ever evolving data volumes we will ever fully mitigate that. Bitcoin for all its merit consumes a lot of energy. Is that more than USD or something else? Another thing to consider is that AI right now has a rich get richer winner take all dynamic. We have titans battling it out for NLP supremacy. Deep mind just revolutionized protein folding which has direct implications into biomedical research and technology. Small academic labs are at an extreme disadvantage unless they collaborate with big tech. Interestingly physics research evolved in a similar way with large particle colliders and telescopes started up. So you can make a direct analogy to that, that the basic research can only be done with million dollar investments. That's an ethical question. GPT-3 trained on misinformation would make pizza gate look tame in comparison. So there are huge social implications. Even real datasets collected from society and social data will have implicit and explicit biases in them. There are huge ethical concerns with data, data collection and with the interpretations AI models will make. Your credit score, your social credit score, a company that refuses to hire you because a chat bot has labeled you subversive. The TSA/DHS blanket banned people from flying and gave no feedback or transparency into why. It took members of Congress to get some people redressed. https://en.wikipedia.org/wiki/Weapons_of_Math_Destruction https://en.wikipedia.org/wiki/Weapons_of_Math_Destruction