A possible example. An AI gets a random goal “Increase intelligence and stop after you reach IQ=200”. It prevents the existence of superintelligences with such goals. So no pure ortogonlaity.
avturchin
Karma: 4,979
I used to think it as following: there are three classes of objects:
1) numbers, like 10101
2) pure qualia, like redness
3) the tables of correspondence which connects numbers and pure qualia. Red is (10101----redness).
Physiological qualia are lines in the table of correspondence, and they combines both functional part (number) and qualitative part (redness).
The number of possible pure qualia is at first glance is very large and maybe unlimited, but the number of tables of correspondence is much larger as each pure qualia can be attached to any number, and also each table includes all colors—this drives combinatorial explosion.
The nature of qualia is relatively simple—they are just a type of mathematical objects which do not depend of anything except themselves. The real problem is the nature of table of correspondence.
Functional part for sure and for qualitative part it is only important that they are different from internal perspective.
The similar question is “Do you think that F=ma represent physical law?”
Moreover, if we take hidden variable in some computation, like x:=x+1, - here x has causal power, but we will not see x in the inputs and outputs.
Qualia are atoms in experience and all complexity is relation between them. Atoms are simple but relations are complex.
After I put this on twitter, I was linked to an article which has the same argument with many details: https://www.academia.edu/22474275/Squalia_Qualia_like_Properties_of_Symbolic_Systems
Only set of possible letters is finite. I think that set of possible colors can be infinite.
Any syllogism which ends with something red, must have something red in its premises. Thus, redness itself can’t be proved in any conclusion. Thus it depends only on itself. The mathematical objects which depends only on themselves are axioms. We can postulate the existence of class “colors” and object red in it.
Red itself is just one of rainbow’s colors and can be encoded in 3 bits. But what you are speaking about is its functional role—its association with emotions etc.
In equation it is more obvious: letter T can denotes one of several physical variables—it an be used for time but also for leghth, energy etc. T as time appears in multiple equations.
If I see something red, there is no much information − 3 bits? - and also it is stable and not shifting.
Qualia are internal variables but they are taken from different realm
I used such three level bullet points style in https://www.lesswrong.com/posts/iBg6AAG72wqyosxAk/the-badness-of-death-in-different-metaethical-theories
It also misrepresent quantum immortality: killing oneself will not move you to better world, but only in worse one—as there are many worlds where suicide will end with severe injury but not death. However. the story can be made stronger if the father of protagonist was trying to build universal lottery winning machine via qunatum immortality.
Also, it is interesting to note that early versions of QI have used time dimension instead of spatial, like in Nietzscheт eternal return.
Yes, but it has strongest predictive power.
I made similar experiments with Nabokov’s story Spring in Fialta. I put first period and asked to continue in Nabokov’s style.
Why you used Chinese models? You can use models with larger context window and put more data about Borges, even Claude Code itself (instructing it never look on the original text of the story). While the result unlikely to be verbatim, a close coincident would mean that you have a good LLM-model of Borges (sideload).
If you fine-tuned open LLM on the story, how you prevent pure memorizing and how you can distinguish memorized output from the original writing?
My point was that prices are dictated by social structure, not by potential abundance of goods. For example, software and movies can be copied unlimitedly, so we have informational abundance, but still have to pay for them—or become pirates and face potential legal risks.
Will Taliban members in Afghanistan enjoy post-scarcity abundance? No.
Will prison inmates get it? No.
Russians? Drug-addicts? Trump-supporters? People who said Y word 10 years ago?
You get it—many groups of people will not have legal access to the abundance. If we list all such groups, likely most people will not get it. And those who will get it, are already rich.
Yes, P doom are meaningless until we have some idea how it can be change. If P doom will have absolutely fixed probability, we can just ignore it.
If we have timing, small changes in it are meaningless, but if it is order of magnitude changes, it has implication on how I spend my remaining life.
Can be useful if we look for genetic material of some late person, but his mother or her daughter is still alive.
One way to convert this probability estimates into something actionable is to convert them into time estimates—how much time we have to find solution for AI Safety. It depends of the shape of probability curve and our lowest acceptable risk estimate.
There is an observation that 10 000 rule’s violation results in 100 near-miss accidents and 1 death (not exact number, just my approximate memory and can vary in different situations). A person can calculate the number of near-misses he survived and calculate if he is affected by survivorship bias.
I used to think that “coherent” here includes that it is also coherent with values of other people, so there is no personal CEVs.
Such universal-CEV may include personal-CEVs but implemented in the simulations without CEV-version of “suffering conscious beings”.