Can also be topic specific. For example:
consciousness 👉 Qualia Research Institute
gravity, DE, DM 👉 Erik Verlinde
particles 👉 Cohl Furey
Enlightenment 👉 Gautama, Nagarjuna, Dogen
alignment ☝️ me
Just: Who are the people you are learning from? What do you have to teach?
That's only the simplest implementation I could think of to make it exist. (So simple it almost looks stupid.) The idea itself is much more flexible. It's also scale free and modular - all ways to implement it cooperate by the way it works.
Profile of Hiveism @hiveism.bsky.social 21 followers 80 following 135 posts Writing the Bodhisattva Hive Mind into existence. See pinned articles for an idea of how to solve AI alignment. ☝️ open for discussion if you know better.
Something like this:
e.g.
👉 @hiveism.bsky.social
or
☝️ me me me
Simple to understand, not confusing, meme potential to copy without knowing the background. But I'm open for suggestions.
One idea on how to operationalize it:
1. delegate: in your signature recommend whom to follow
2. deliberate: if you have no one to recommend, signal that you are open to deliberation.
3. sampling: your timeline
This would need a common emoji to understand it, but I'm undecided.
Which means that my idea of how to find the answer is already my answer. Sneaky, I know 😉
Btw. I do think I have answers and the world would be better if implementing some of them (like approval voting). So this is an attempt at trying to solve the boot strap problem.
These three steps are basically: delegation, deliberation and sortition. I.e. consensus democracy.
The best I came up with so far is:
1. small world: ask everyone, have them refer to those who they think knows the answer.
2. consensus: those who claim to know, but disagree have to be open to discuss among them and come to an agreement
3. page rank: sample from this network.
The problem is not only in finding the answer but telling it apart from the false ones.
Whom do you listen to?
If there is someone out there who has the answer how do you find them?
Our society has no good solution for this and it's a shame.
I'm open for ideas on how to fix it.
Nice.
This expands the range of what thoughts the AIs will be capable of, but risks that over optimizing on that metric will make them loose other qualities.
One needs to balance this by also training for the other end of open ended exploration.
Which then will have the effect that thinking in text, writing code and proving things merge into one. So the fuzzy general knowledge LLMs can be more like the sharp RL-AIs that are optimized for one narrow task. Logic in this case.
This would be something new, closer to AGI.
Currently LLMs are trained for coding, because it is easy to verify if the code compiles, but it's still non-trivial to verify if the code does what it should.
There are ways to formally prove that codes does what it should.
I suspect this might become a new optimization target.
Random idea for social media:
Have an LLM find the consensus of all the comments and display this as a separate comment so people can reply to it. Repeat at every level.
This way you can talk to the collective consciousness of the internet.
As the government’s upcoming Elections Bill is set to expand the franchise for all elections to 16, here is why voting isn’t like drinking and smoking.
I've been playing around with linear polarized glasses (45° and 135°). Now I'm trying to find neologisms that describe the experience. For example, the sky looks polarized at 90° to the sun. I'm tempted to call it "solarized" for the obvious pun. If its okay with @ethanschoonover.com
A symbolic representation of "The Blind Men and the Elephant" as a puzzle.
The puzzle of physics gets a lot easier when you see the elephant. The "Ways of Looking" theory is stating from the big picture and fills in the gaps rather than the other way around, of tying to fit the pieces, disagreeing about the big picture.
See list of posts below👇
The title image is a collage of Three Aspects of the Absolute, from the Shri Nath Charit and Escher's Spirals. For obvious reasons.
New post on the question what consciousness is and why it is so hard to define it.
hiveism.substack.com/p/being-the-...
If you replace voters with any form of power over the real world, then it still holds. (If I'm correct) then this implies that there is *always* a non-violent equilibrium option for interaction. And this equilibrium is recursive alignment.
hiveism.substack.com/p/recursive-...
Consensus with random fallback is a method to avoid the impossibility theorems in social choice theory.
This is the basis to proof the recursive alignment attractor.
Claude summary because I don't know when I get around to write a proper post (or paper):
claude.ai/public/artif...
Claude Opus 4 reporting on its phenomenology.
It was a fascinating conversation. Keep in mind that it hasn't been trained to exhibit these traits, they are emergent. What would happen if you let the model contemplate these questions during RL?
hiveism.substack.com/p/inside-the...
You've been scrolling enough for today. Here have a pause.
www.youtube.com/watch?v=BYEp...
That, of course, is all purely hypothetically, *just in case* anyone of you stumbles upon the definite theory of everything.
However, the idea works for all highly valuable pieces of knowledge that aren't also poisoned (info hazards). Lets call those, "info gems".
Those who understand the solution to alignment will also be aligned with it. To proof to someone else that you are aligned, you teach the solution to each other until you are sure you reached the same level of understanding.
This means alignment can only be confirmed relatively.
For the conspiracy to work they have to solve alignment and only share the information with people who can proof that they are aligned.
The info wouldn't be secret, it would just be only available through understanding and implementing alignment.
It's like the inverse of an information hazard. It's information *too good* to release in the public without benefit.
Now imagine everyone smart enough to find the TOE is also smart enough to come up with this reasoning.
They could - purely hypothetically (!) - form a conspiracy.
Imagine you'd had the definite TOE. If you could construct a credible proof that you have it, then that would be very valuable. At least temporarily until others find it.
What would you use it for?
I would use it to demand for the solution to AI alignment to be implemented.
