I think floating point is viable - balatro doesn't exactly perform billions of floating point operations per second.

I think it's simpler than that. I think they hire people interested in the subject of AI safety and give them relatively free hands to publish what they find, and findings don't necessarily have to be part of some agenda that benefits Anthropic.

The benefit instead comes from having these competent passionate people employed and their knowledge somehow contributing to better and safer models.

This but unironically. In the movie no one was killed for the purpose of being turned into soylent green, they just used bodies of people who were dead anyway. I remember thinking that it's not such a big deal compared to all the other misery in that movie, and that ignorance was bliss in that respect.

I don't fully remember the movie anymore, but there were many other things that were much worse that actually happened on an ongoing basis and leading to people suffering. Like the poverty.

Sure but you can ask that at different scales, in a reductio ad absurdum: Why should EU use American tech company? But also why should Germany use French tech company? Why should one region of France use tech company from other region? Why should one person use tech from another person?

Would addition be slower than for normal floats?