In one camp, we have people at organizations like Microsoft/OpenAI, Google, Facebook, and maybe Baidu that have successfully trained large-scale AI systems with

In one camp, we have people at organizations like Microsoft/OpenAI, Google, Facebook, and maybe Baidu that have successfully trained large-scale AI systems with hundreds of billions to trillions of parameters. People in this camp say they are worried about societal risks, but they keep moving forward nonetheless. I wonder if it's because to them the rapid improvement in AI capabilities looks like a shiny rainbow with a big "pot of gold" (money, fame, glory, etc.) at the end. In the other camp, we have people at other organizations, including every academic institution, who cannot or so far have not been able to train a large-scale AI system with hundreds of billions to trillions of parameters. People in this camp are the ones writing open letters about societal risks. These people sound sensible, but I wonder if they're worried because the rainbow is protected by a giant wall, and they see themselves as being outside that wall (e.g., they lack the budget or expertise).