Get live statistics and analysis of Ashley Darkstone's profile on X / Twitter

I think I'm refining my design into something that goes beyond neuro-symbolic. Something kind of new...

Progress Update: Figured out the technical implementation. Building a demo of the core cognitive functions. This thing is going to be very very lightweight!

The heart of my engine... -A new type of "reasoning database" -A new form of attention -Neuro-symbolic nets I may release the database as a standalone product.🙂

Neural nets aren't neurons, you aren't building a brain. Machine learning is an evolutionary dead-end.

An elegant AI architecture for a more civilized age.

It's officially been 9 years since I started working on my cognition engine. Crazy. Went from inspiration.. just a vision of processes I didn't fully understand, to a bunch of bloated diagrams, to a refined vision and understanding. And now? A technical implementation to actually realize my vision. Not just that, but it's so much more pragmatic and lightweight than I ever imagined, thanks to a breakthrough earlier this year. I'm very close now. Aiming to have a novel database, a core component of this engine, completed by next year, and over the next 2-3 years I hope to build a full prototype engine. It will solve most or all problems with ML and symbolic AI (and is technically a hybrid neuro-symbolic system). I just wish I could get it done sooner, so we don't have to suffer through LLMs any longer...

I can't not build AGI. I'm burdened by knowledge.

I think I finally figured out what specific form passive "subconscious" background learning will take within my cognition engine. 😃 Was a bit worried about resource usage but the architectural principles of my design make it lightweight and modular like everything else.

Seems that if I strip out a lot of parts from my engine, it would only go from JARVIS to LCARS, possibly MUTHUR. Not a bad downgrade... The biggest difference is proactive vs. reactive.

One of my best features is my fusiform gyrus.❤️

Working on dynamic weighting today, and planning a simple proof of concept of the core cognitive functions I can get done sooner rather than later.

I think I just came up with a method for a neuro-symbolic net for my cognition engine that will map 2D image transformations to 3D, and track them in real-time. I love this shit. 😁 But it'll have to wait, I'm doing language first, then images.

Status Update on the development of my 'General Cognition Engine': -Planning communication and process flow to align with the technical implementation. -Figured out my version of the tokenizer (it's not tokens though, this isn't an ML system). -Figured out how the attention mechanism will work with the technical implementation, and how to accomplish all the attention types. -Have a method for introducing fuzziness across the system. -I know how my neuro-symbolic nets operate conceptually, and aside from a few details, technically as well. A partial mystery is the synthesis part, which relies on other processes, so it'll become more solid once I determine the flow of everything else. -Figured out how the database will function both conceptually and technically, and how it integrates with the attention mechanism. -Figured out my data types, how the data is structured, what language I'm using, and what backend software I'm using. -I mostly know how I'm handling learning, generalization, context, planning, predictive analytics, self-awareness, and more. Very confident in this direction, and I'm already starting to build the uhh 'main manager', 'tokenizer-equivalent', data intake process, and the database.

Few know this about me, but I want to build Starfleet starships... for real. Who is building the tech to make it possible? Anyone working on radiation shielding for space travel? Faster-than-light drives? Photonic computing? Inertial damping? Advanced materials? Artificial gravity? I'll bring an AI operating system to the table. Almighty algorithm, where are the sci-fi tech nerds building cool shit that brings us closer to a Star Trek future?

My AI architecture is almost TOO flexible..

Neural nets aren't neurons, you aren't building a brain. Machine learning is an evolutionary dead-end.

My AI architecture is almost TOO flexible..

VCs need to stop exploiting children.

In designing my Cognition Engine, I invented a new form of modular neural net and attention mechanism. While it's not meant to learn/train like the Transformer does, it could be adapted to proven ML methods, while still being a lightweight and modular 'white box'. So even if my learning/training methods don't pan out, I could still win.🙂

It's officially been 9 years since I started working on my cognition engine. Crazy. Went from inspiration.. just a vision of processes I didn't fully understand, to a bunch of bloated diagrams, to a refined vision and understanding. And now? A technical implementation to actually realize my vision. Not just that, but it's so much more pragmatic and lightweight than I ever imagined, thanks to a breakthrough earlier this year. I'm very close now. Aiming to have a novel database, a core component of this engine, completed by next year, and over the next 2-3 years I hope to build a full prototype engine. It will solve most or all problems with ML and symbolic AI (and is technically a hybrid neuro-symbolic system). I just wish I could get it done sooner, so we don't have to suffer through LLMs any longer...

The saxophone is the best musical instrument man has devised thus far. That is all.

I can't not build AGI. I'm burdened by knowledge.

I think I'm refining my design into something that goes beyond neuro-symbolic. Something kind of new...

I think I finally figured out what specific form passive "subconscious" background learning will take within my cognition engine. 😃 Was a bit worried about resource usage but the architectural principles of my design make it lightweight and modular like everything else.

Seems that if I strip out a lot of parts from my engine, it would only go from JARVIS to LCARS, possibly MUTHUR. Not a bad downgrade... The biggest difference is proactive vs. reactive.

The heart of my engine... -A new type of "reasoning database" -A new form of attention -Neuro-symbolic nets I may release the database as a standalone product.🙂

Ended up working on sensory and data fusion yesterday. Pretty sure I know how I'm handling that. Things are getting easier now that I know what I'm doing. My basic neuro-symbolic net design is highly modular and scalable, and the technical implementation is fairly simple. Makes it easier to design new 'nets, though the 'basic' structure already handles a lot of types of processes without significant modification.

One of my best features is my fusiform gyrus.❤️

Working on dynamic weighting today, and planning a simple proof of concept of the core cognitive functions I can get done sooner rather than later.

I think I just came up with a method for a neuro-symbolic net for my cognition engine that will map 2D image transformations to 3D, and track them in real-time. I love this shit. 😁 But it'll have to wait, I'm doing language first, then images.