If there are no real worlds, but only BBs all along, this argument doesn’t work.
However, it is still not a big problem, as Dust theory still works, and for any BB there will be another BB which represent its next mental state. So from inside it will look like normal world. Mueller wrote a mathematical formalism for this.
It was not google, but Microsoft: in September 2020 they wrote: “The trillion-parameter model has 298 layers of Transformers with a hidden dimension of 17,408 and is trained with sequence length 2,048 and batch size 2,048.” https://www.microsoft.com/en-us/research/blog/deepspeed-extreme-scale-model-training-for-everyone/
We miss GPT-4 data point for this chart. New Megatron is more like one more replication of GPT-3, and another one is new Chinese 245 Billion parameters model. But Google had trillion parameter model in fall 2020.
I read a pdf some time ago (link lost) where was listed number of deaths for different types of cars. I remember that Kia Rio has 1 death for every 10 mln km, Toyota Prius − 1 for 1 billion and Dodge Caravan had no deaths. However, later I found similar statistic and Prius had more accidents.
Am I correctly understanding that: If we use exponential as a universal prior, the simplest hypothesis is 2 times more probable than the nearest hypothesis which is 1 bit longer?
And that the simplest hypothesis has equal probability than all other longer hypothesis combined?
If 99.99 per cent of all civilizations go extinct, SIA Doomsday is true in basement reality, despite the fact that last 0.01 civilizations create trillions of simulations, and we are likely to be in one of them.
Interestingly, if we are in a simulation, it still could simulate the end of the world, and the share of such simulations should be large by the reasons I explain below. Thus on the observable level, SIA Doomsday still holds, despite the fact we are likely in simulation.
Th reason for the large number of “Doomsday simulations” is that any aliens will try to solve Fermi paradox numerically, as they want to learn what is the probability to meet other aliens. So they will simulate many civilisation’s histories around the time of high x-risks (that is 20-21 century in our world). Not only their own past but the history of any other possible civilisations. So we could be simulated by aliens which explore the ways how the worlds tend to end.
If there is no other reason to create simulations, such Fermi-paradox-solving-simulations will be most numerous. Other reasons, like gaming, may also include the end of the world as a bad out come in “world saving game”.
This is rather obvious: I am randomly selected from all beings how can think about reference classes. Ants can’t think about that. Even most humans before 20 century were not able to think on this topic as the needed ideas were not known.
Check this: “Mammalian Value Systems” https://arxiv.org/abs/1607.08289
What do you think of the role of modal realism in SIA vs SSA discussion?
I mean that if everything possible does exist, then any SIA situation is actually SSA situation. For example, if some process (God, or panspermia) creates a million copies of you in one region of the multiverse, there always will be another region, where this process failed and created only one your copy. Presumptuous philosopher is now telling that you just more likely to be in the region of multiverse where there are many your copies.
Yes. I found that “male life expectancy at birth is 76 years. That’s the mean value. The median life expectancy is just past age 80. And the mode (i..e, most common) age at death is age 86.”
When you say “the year of PASTA”, you probably mean the year than AI appears with 50 per cent probability. But why “50 per cent probability”? 10 per cent seems to be more important. For example, when we say “human life expectancy is 75 years”, it means that in the half of the worlds I will die before 75. The same way, by using the median year as a measure of AI timing, you already accept the loss of the half of human future when AI will appear before that date.
More generally, speaking about the “year of AI” is meaningful only if the dispersion of the Probability-of-AI-appearance(t) is small. If 10 per cent is 2030, 50 per cent is in 2100 and 90 per cent is in the year 3000, than saying that AI will appear in 2100 is a completely misleading picture.
That is, there are two problem in using year as a way to estimate AI-timing: 1) humanity will go extinct in the half of cases before this year 2) it creates a false impression that AI probability of appearance is a bell-like curve with a small deviation from the mean.
We could increase chances of this by making commitment to run in testing simulations many copies of different possible Rogue AIs after we create a Friendly AI. This is an idea by Rolf Nelson.
Moreover, as some Rogue AIs will try to emulate Friendly AI, they will take this commitment for granted and simulate other possible Rogue AIs in a nested simulation. So it becomes self-fulfilling prophecy.
I found that GPT-generated porn is better than what human can typically write. It has perfect style when it is writing about sensible topics, without any internal hesitation or over-expression.
Yes, exactly this. I am working on a text on personal identity, and come to similar conclusion
Life-logging and self-description
I like this idea. Most of these probabilities become actionable if one thinks what will be after death (and is suicide good?). Hell? Another level of simulation? Quantum immortality? Future AI will create my copy? Nothingness?
Answers on these questions depends on one’s metaphysical assumptions. If he has a probability distribution over the field of possible metaphysical ideas, he may choose best course of action reading death.
Actually, I am not emotionally disturbed. But my immortality hopes shifted to indirect digital immortality from cryonics.
Actually I heard that she lost a lot of money when she was cheated by Italian criminals.
In his article Mueller says that no physics exists at all. Only math world exists, and dust minds are just random strings of digits.