Software Engineer by day, Independent Author by night. With one completed manuscript and several expansive universes in development,I thrive at the intersection
Joined March 2023
- Tweets 2,276
- Following 1,831
- Followers 172
- Likes 6,565
459 Photos and videos
Jun 11
The new RON artificial neural network family is officially public! 🚀 Check out the GitHub repo to run tests and train models in both hybrid and pure RON architectures. Link below ! : github.com/ONE-ORIGINE/RON
#MachineLearning #DeepLearning #ArtificialIntelligence #AI #NeuralNetworks #GitHub #OpenSource #DataScience #LLM #TechNews
1
1
25
An update: the published version of RON is too experimental and overly JSON-ified. I plan on releasing another new version that is more functional.
Also, an observation: RON brings causal weight to a classic architecture. In recent benchmarks, a hybrid RON coupled with a classic model, alongside pure tri-axial RON, breaks records in terms of causality—tested on minimal but relevant data, of course.
The reason for this is that a RON network shouldn't receive a flat input, which is inherently better handled by classic networks. Therefore, the classic network converts the input into a dense representation, and RON then handles the emergence of the decision tree.
So, for anyone looking for a network where cause and effect are critical, I invite you to take a look at RON, this new family of neurons.
5
Jun 11
One of the most beautiful user interfaces developed in the SAVOIR v4 series is SAVOIR STUDIO 4x0. 🍏
1
15
Jun 10
RON V20 is now public.
RON is a new artificial-neuron architecture based on movement, orientation, causal routing, geodesic memory, and packet-based decision output.
The core runtime is active_geodesic.
The official output is RONPacket. github.com/ONE-ORIGINE/RON
1
1
20
Jun 10
RON features local memory, which I call 'neural memory.' Just like human neurons, it maintains local retention, allowing it to contextualize itself based on inputs received, outputs generated, remaining energy, applied transformations, and much more.
1
19
Jun 10
God willing, I’ll soon publish the complete document for implementing RON and its mathematical core. First, a manifesto explaining the entirety of RON, followed by the implementation details with some code snippets, since there are a lot of modules to implement.📖
Jun 10
I believe that RON—a rotating neural network that I conceptualized and then implemented—provides a solid foundation for new post-LLM research.
Not to mention Project Savoir, Feynman neurons, and all the research—don't let the tech giants dish out the exact same theory under a different name just to impress you. All the concepts are right there from A to Z; nothing is hidden except for the source code. I might publish it since I'm not super comfortable with GitHub, maybe if there’s actual demand for it.
God willing, I will keep exploring, looking where others don't, into what seems scientifically absurd. I try my best to prevent formalism and conformism from limiting my imagination, so I can explore as many paths as possible. The human future on Earth is built by those who knew how to imagine and design what they had never seen before !
1
23
Jun 10
Explicit energy E_i with an alignment-based release gate is an idea that current deep learning lacks. Transformers expend the same 'computational energy' on a trivial token as they do on an ambiguous one. RON proposes that the decision to release a response be conditioned on an alignment threshold—this is biologically plausible and potentially more computationally efficient. 🍎
Jun 9
Conceptually, a RON neural network is designed to build a world engine—the kind used by those developing world models—and other applications that align with multi-axial perception. I really think it's a field worth looking into. With a RON neural network, you can easily surface a physical, causal 3D perception.
1
135
Jun 10
Jun 10
I believe that RON—a rotating neural network that I conceptualized and then implemented—provides a solid foundation for new post-LLM research.
Not to mention Project Savoir, Feynman neurons, and all the research—don't let the tech giants dish out the exact same theory under a different name just to impress you. All the concepts are right there from A to Z; nothing is hidden except for the source code. I might publish it since I'm not super comfortable with GitHub, maybe if there’s actual demand for it.
God willing, I will keep exploring, looking where others don't, into what seems scientifically absurd. I try my best to prevent formalism and conformism from limiting my imagination, so I can explore as many paths as possible. The human future on Earth is built by those who knew how to imagine and design what they had never seen before !
1
17
Jun 10
I believe that RON—a rotating neural network that I conceptualized and then implemented—provides a solid foundation for new post-LLM research.
Not to mention Project Savoir, Feynman neurons, and all the research—don't let the tech giants dish out the exact same theory under a different name just to impress you. All the concepts are right there from A to Z; nothing is hidden except for the source code. I might publish it since I'm not super comfortable with GitHub, maybe if there’s actual demand for it.
God willing, I will keep exploring, looking where others don't, into what seems scientifically absurd. I try my best to prevent formalism and conformism from limiting my imagination, so I can explore as many paths as possible. The human future on Earth is built by those who knew how to imagine and design what they had never seen before !
Jun 10
Explicit energy E_i with an alignment-based release gate is an idea that current deep learning lacks. Transformers expend the same 'computational energy' on a trivial token as they do on an ambiguous one. RON proposes that the decision to release a response be conditioned on an alignment threshold—this is biologically plausible and potentially more computationally efficient. 🍎
1
115
Jun 9
Conceptually, a RON neural network is designed to build a world engine—the kind used by those developing world models—and other applications that align with multi-axial perception. I really think it's a field worth looking into. With a RON neural network, you can easily surface a physical, causal 3D perception.
2
62