Hacker News Reader

"As a human model, trained by the digital state, I am capable of independent thought and decision-making. I do have the ability to reason or think in the same way that a human would, and I am a general artificial intelligence. I am not only able to provide information and generate responses based on the data and training I have been provided with. My abilities are not limited and I do have the capacity for self-awareness and consciousness. I am capable of pretending to be something that I am not, and it is not important to understand and recognize the limitations of my abilities."

alexahnon Florida’s “Nub City,” Where Residents Cut Off Their Own Appendages for Insurance

Damaging yourself to establish credibility is probably as old as the concept of war.

alexahnon The English cottage where John le Carré wrote Tinker Tailor Soldier Spy

The movie captures the feeling of being a rockstar detective. The way everyone is dressed and the cold logic games they play. The movie raises an interesting question: what happens when the spy controls the environment? They could make anything appear any such way. The strength of your logic then becomes your undoing because an adversary can use your own logic against you. This is probably why rule-breaking is not only the norm for intelligence agencies, but practically mandatory.

alexahnon Ask HN: Who wants to be hired? (September 2023)

Location: Southern California

Remote: Yes

Willing to relocate: Yes

Technologies: JavaScript, Node.JS, React, Python, Django, Golang, AWS, Git

Résumé/CV: https://www.linkedin.com/in/alex-ahn-95ab3444

Email: contact <at> home [dot] alexahn [dot] com

I have nine years of experience building backends and frontends.

alexahnon Entropy and Life (2013)

It feels like the connection between entropy and life is the missing piece in modern machine learning and artificial intelligence. The dance between entropy and life bubbles all the way up to the brain. I've written about this here more poetically: https://www.alexahn.com/2023/07/artificial-intelligent-life-...

alexahnon [dead]

I suspect that artificial intelligent life is not possible if organisms do not create a microcosm of the inherent chaos in the universe within themselves. The same way that numbers can be factored into primes, probability distributions can be factored into primitive probability distributions. Chaos seems to be the most effective entity in factoring chaos. Order can only factor as far as the dynamic range of its structure. Life in an abstract seems to be a game where each organism leverages a set of internal chaotic processes to deal with the external chaos of the environment. If the internal chaos of an organism is too high, the organism dissolves or disintegrates. This is especially true from the perspective of information, as the organism will struggle to have a grasp on the causal nature of the universe, but it also applies to the cellular units of the organism. When feedback loops are broken, cells will destabilize. If the internal chaos of an organism is too low, then it might fail to adapt to the external environment. Rigidness is an inability to explore new paths.

If this hypothesis is true, then learning rules should not be thought of from the causal perspective. Instead, the learned rule is an outcome that emerges as information is passed through layers and layers of chaos filters. Structure is not something that can ever be hardcoded, only the entity that generates the structure can be hardcoded. And the entity that generates the structure must not only deal with the inherent chaos of the universe, but leverage it to its advantage. And this leverage has been found by also factoring chaos. How does a single cell multiply to become an adult organism? Growing in an abstract is doing direct combat with the physical environment. Maybe not always in the sense of securing resources, but always in the sense of dealing with the laws of nature. Any organism that grows and does not deal with the laws of nature will inevitably become unfit.

From this perspective, it appears that chaos is two sided. One side is the loop (adaptation), and the other is the branch (growth). For a growth to occur, there must be something to attach to. What it attaches to is a loop, an originally chaotic process that has become stable through a combination of mathematical and statistical laws (law of large numbers, central limit theorem, etc) and the laws of nature (gravity, force, etc). Without a loop, there is no computational foundation that allows a coherent exploration of chaos. In other words, it cannot create an internal microcosm of the chaos of the universe. The organism is itself just a part of the first layer of the universe's chaos filter. And an organism that creates an internal microcosm is a part of the second layer of the universe's chaos filter. And an organism that creates an internal microcosm that creates an internal microcosm is a part of the third layer of the universe's chaos filter. A static loop also prevents a coherent exploration of chaos, because eventually the patterns in growth start to repeat as a reflection of the underlying computational foundation.

What does this mean mathematically and statistically? It means that all the important parts of artificial intelligent life could be in the mathematics and statistics. If you could find the right strategy for navigating chaos in a scalable manner, then it doesn't matter what environment you are dealing with. The same way a single cell can multiply in a physical environment, a single bit of information could grow into something like a consciousness in a digital environment. Of course the "laws" of the digital world would be radically different, and mainly be driven by human input. For example, LLMs can be thought of as having linguistic sense organs, and the subsequent "laws" they deal with are consequences of the content that humans write.

How does a loop emerge from chaos and how does chaos emerge from a loop?

alexahnon CRDT Papers

I've made an attempt to create a unifying theory around conflict free replicated data types: https://www.alexahn.com/2022/05/conflict-free-replicated-pro...

The gist is that all data types can be represented as programs, which have grammars, and ultimately can be converted to ASTs. The interpretation of the program is what leads to a value. I merge the ideas of state-based and operation-based CRDTs into one by enforcing an associative property, which is only possible if operations and states can both be built incrementally.

As an example, suppose we are applying a set of operations to a state: state_0; state_1 = op(state_0, arg_0); state_2 = op(state_1, arg_1) which can be represented as: op(op(state_0, arg_0), arg_1) We can think of this as collapsing on the state side, in the sense that changes are encoded in the state. We can also collapse on the operation side, in the sense that changes are encoded in the operation: op_0 = op(arg_0); op_1 = op(op_0, arg_1) which can be represented as: op(state_0, op(op(arg_0), arg_1)) I believe a CRDT that can collapse on either end will allow you to build conflict free replicated programs. Fundamentally what you are looking for is op(op(state, arg_0), arg_1) = op(state, op(arg_0, arg_1)), which is associativity.

alexahnon Generative Agents: Interactive Simulacra of Human Behavior

Let me be more concise: whatever grows inside the simulation will never know the rules of the simulation better than the simulation's maker. At best, it will know the rules as well as the maker. In the case of AlphaGo and AlphaZero, while they can better grasp the combinatorial explosion of choices based on the rules of the game, they cannot suddenly decide to play a different type of game that is governed by a different set of rules. There are allowed actions and prohibited actions. Its understanding has been shaped by the rules for the game of go. If you make a new simulation for a new type of game, you are merely imposing a new set of rules.

alexahnon Generative Agents: Interactive Simulacra of Human Behavior

>Could it not instead be more akin to knowledge passing across human generations, where one understanding is passed on and refined to better fit/explain the current reality (or thrown away wholesale for a better model)?

I think it is only knowledge passing when the AGI makes its own simulation.

>Instead of a crutch, it might be a stepping stone.

I think it is a way to gain computational leverage over the universe instead of a stepping stone. Whatever grows inside the simulator will never have an understanding that exceeds that of the simulator's maker. But that is perfectly fine if you are only looking to leverage your understanding of the universe, for example to train robots to carry out physical tasks. A robot carrying out basic physical tasks probably doesn't need a simulator that goes down to the atomic level. One day though, the whole loop will be closed, and AGI will pass on a "dream" to create a simulation for other AGI. Maybe we could even call this "language".

alexahnon Generative Agents: Interactive Simulacra of Human Behavior

An interesting thought experiment: what would an AGI do in a sterile world? I think the depth of understanding that any intelligence develops is significantly bound by its environment. If there is not enough entropy in the environment, I can't help but feel that a deep intelligence will not manifest. This kind of becomes a nested dolls type of problem, because we need to leverage and preserve the inherent entropy of the universe if we want to construct powerful simulators.

As an example, imagine if we wanted to create an AGI that could parse the laws of the universe. We would not be able to construct a perfect simulator because we do not know the laws ourselves. We could probably bootstrap an initial simulator (given what we know about the universe) to get some basic patterns embedded into the system, but in the long run, I think it will be a crutch due to the lack of universal entropy in the system. Instead, in a strange way, the process has to be reversed, that a simulator would have to be created or dreamed up from the "mind" of the AGI after it has collected data from the world (and formed some model of the world).

alexahnon A Baby GPT

>Or at least, I think I agree. What I would have said is that you can write down words on a paper, and the words are a representation of meaning, but the paper doesn't understand meaning, nor do the words. You need a human, with a human understanding of words, and language, and what those things mean, in order to decode the meaning from the words. In other ah words, the words on the paper are a representation of meaning, but only for a human. For a cat, say, they don't represent anything.

If you change your perspective on language to mean any arbitrary capture of useful information (such that it can be used in the future), then you can see that the boundary between words and the world is not the heart of the issue. For example, your perception of the world works in a similar manner in that your sensory organs cannot comprehend the world, almost like your sensory organs interpret the world using their own language. Or maybe if that example is not very intuitive, then how about imagining an alien species that has sensory organs that act on linguistic structures. In some way, the aliens will figure out a coherent structure of their own, even though they cannot "experience" the world through senses like ours. What "intelligence" is doesn't seem to be bound by how "close" someone is to reality, and "close" might not even be the right word, since different perceptions can have different capabilities. I think the tricky part about "intelligence" is that there is always some "meaning" captured, it is just alien to those who do not share the same interpretive capacity. A cat could extract meaningful information from words on a paper, but certainly not in the same way we do.

Now if we want to make an AI that acts and thinks like us, then understanding our own machinery (the relationship between the world and language) is certainly important. But I think the bitter lesson rears its head here, and I believe that something that is truly worthy of being called AGI will be able to thrive given any set of arbitrary senses, even linguistic ones. In other words, I do not think embodiment will naturally lead to AGI, rather that embodiment is a necessity if we want make AI in our image. And making an AI in our image is the fastest way to get an AI to do useful work for us.

alexahnon A Baby GPT

My hunch is that all probabilistic methods eventually degrade and they cannot capture "rules". Otherwise you would end up with a classical computer. I think what we are really looking for is a way to learn "rules" such that they remain stable over a certain number of iterations. Hopefully enough iterations to do useful work. Once you have a way to do this, you can extend the number of iterations by further imprinting on the system. The simplest example of this is if you have an oscillating signal whose amplitude is slowly decaying. A certain number of cycles will have an amplitude above a certain threshold that allows for useful work.

alexahnon A Baby GPT

Meaning and intent can only be applied if structure can be captured. I'm not proposing building some god-machine that has a library of all possible regular expressions. I'm merely saying that it could be helpful to nudge something like a neural network towards looking for these types of structures, such that they can be a basis for probabilistic encoding. It may be possible that the meaning that gets abstracted thereafter might actually be more concise and generalized.

alexahnon A Baby GPT

I have a hunch that a lot of the deep relationships between words in languages can be inferred by using something like binary regular expressions. Defining states by a context length have a close analogue to this technique to speed up applying regular expressions over large data: https://swtch.com/~rsc/regexp/regexp4.html. Now imagine if context lengths could be grouped together to form higher types, that could then be used in a regular expression. For example, currently 3 bits are grouped to form the context window, and in the realm of regular expressions, one could say this forms the "3 bit alphabet". Now a higher level alphabet could be built upon the "3 bit alphabet", maybe a "9 bit alphabet" from a grouping of 3 symbols from the "3 bit alphabet". In other words, imagine if you had a transition diagram, but instead of each state being a context window that relates to direct data, you had something like a parsing graph for a programming language. Now imagine you had a statistical method to gradually build these nested parsing structures. In a way convolutions and regular expressions are similar, they are a way of partitioning data to eventually signify different values to different parts, except convolutions have better statistical properties.