@EvanHarper I mean, if Russia didn't have nuclear weapons, Polish troops would be in Moscow right now, so I think they're still pretty useful.

@rexcirus What I'm concerned about is that in AI, such distinctions aren't as clear cut. Would you say that AlphaZero wasn't a 'single AI' but rather MCTS and a neural network stuck together? Why couldn't you think of multiple neural networks glued together with some conditional statements as a 'single' AI, since it behaves the same as a 'general AI' when looked at from the outside?

It might be good to include a minimum amount of training flops as well in a question like this. I expect the first 100T parameter model to be severely under-trained for its size.

@mulliken I doubt that's right. Context window is way too small to fit the entire novel, and even if you fine-tuned the model on the book's text, I don't think it would remember long term narrative patterns.

@rexcirus What's the difference between a general AI and 4 narrow AIs glued together? A single neural network?

@EvanHarper Hmm, yeah, the counterfactual isn't obvious. If Russia didn't have any nuclear weapons, Putin probably wouldn't have invaded Ukraine in the first place since NATO would have had little qualm with getting directly involved. If Putin would have invaded anyway though, it isn't clear if NATO would have just kicked Russia out of Ukraine or tried to bring about regime change.

I'd be curious to hear people's breakdown here. What are the odds of each of the 4 options in the US did/didn't do it and will a NATO country investigation find that they did it by January 1 2025 quadrants.

I would be interested to see a question like, will the company currently known as Twitter have a valuation > $100B before 2030. It doesn't seem impossible that after a period of chaos, Musk manages to turn it around. He does seem to have some big ambitions for it. I'd give it ~15%.

@perdana I guess that if it happens, it should be at the Tesla AI event at the end of the month, so until then, I don't really see the lack of an announcement/demo as material evidence either way.

My mode outcome is that sometime in 2024 or 2025 a chip manufacturer or super-computer operator announces the training of a >100T transformer model with a performance slightly above GPT-3, mostly for the sake of demonstration.

@Jgalt: "Don't confuse your prediction with my resolution."

4% chance that this question resolves as ambiguous.

@andrewallen1.6180 I think it should resolve positively. But a lot of what Musk has been doing seems like it could also have been done by a company board of directors. Not an expert in corporate governance, but the line between business owner telling executives what to do, and actually being CEO, seems a bit blurry.

— edited by leon.waterman

@perdana Why should it change either way specifically? Do you take the absence of any early demo/leak as evidence against it happening?