What can it do?
Online Learning
She can browse the web, use that to build databases, and/or fine-tune itself. It is constantly doing so.
Natural Speech
It has conversational text to speech capabilities, and can clone any voice.
Multi-modality
X41A can comprehend image, voice, and video inputs.
Image Generation
Through a combination of Mothership and Stable Diffusion, with a large library of LoRAs.
Software Synthesis
X41A can create its own tools through Mothership, Sandboxing, and NEAT.
Mothership provides very strict, dynamically generated code that is guaranteed to run; this system can tackle the creation of many different small tools in a novel way that is very different from human coding heuristics.
Sandboxing is coding like a Language Model does. It creates a container to which it downloads dependencies and sets up an API. It then iterates on compiler feedback against a series of Tests. As of right now, its main programming language is Rust.
NEAT stands for Neuro Evolution of Augmenting Topologies. It is a way of exploring both topology and weights of a Neural Network at the same time. Excellent for video-game playing.
Augmented AI Chat
Including RAG, Knowledge Graphs and Tool Usage.
Since she runs on an ensemble of language models, its context size is dynamic and can vary from each internal iteration to the other.
As of right now, X41A has to be asked to stop doing whatever it is doing, and then it can allot some time to talk in its videogame-like world.
Trading
Combining Mothership's battle tested trading abilities with X41A's adaptability and self-learning.
Trading and Research are its main concerns at the moment.
Optimization
Given the right constraints, it will generatively iterate a series of designs—in multiple domains—until hitting the right targets.
The GOLEM can write its own simulations/tools that it wants to see optimized. This is its key defining feature.