Systems & builds

An indexed history of the AI and AGI projects, their architecture, source branches, equations, and evidence.

292 recordsSource snapshot 2026-10-17

292 records · page 3 of 6Search the source archive ↗

github-original-thataiguy

ThatAIGuy original · MB1 system

The original ThatAIGuy tree coordinates a five-layer, twenty-generator tick architecture, continuous learning, seventeen Pantheon names and federation. Those counts come from this generation’s README and configuration; later rosters are preserved separately.

System recordSource recovered; static review

github-original-seferim

Seferim original · MB2 system

The original Seferim tree combines external knowledge collection with named G16 state records and several batch/autonomous learner variants. Its source and tracked state should be distinguished from the later C++ G16 scaffold and JavaScript mathematical reference.

System recordSource recovered; static review

github-unified-v3

Unified AGI Core v3.0 · Synthesis

This synthesis generation explicitly combines the original ThatAIGuy and Seferim architectures. Its README configures an 8,000-node substrate, thirty-two generators, seven layers and a twenty-four-name Pantheon; these are source-defined architecture claims rather than measured capacity.

System recordSource recovered; static review

github-mind-farm

Mind Farm · integrated C++ research system

The recovered branch adds a distinct C++ system with Gate5000, neurological equations, DNA dynamics, eigenvalue operations, learning mathematics, attention, knowledge ingestion, persistence and a unified tick loop. Its headers expose the component-level formulas and the intended interaction of…

System recordSource recovered; static review

github-generator-core

Generator Core · Go generator framework

This branch implements a Go framework whose registry allocates one hundred generator IDs across ten families: autoregressive, diffusion, adversarial, variational, flow, energy, transformer, recurrent, state-space and mixture. It adds attention, DNA dynamics, consciousness-named metrics,…

System recordSource recovered; static review

github-unified-cpp-consciousness

Unified C++ Consciousness Core

A separate consciousness_core.hpp connects the expanded Unified orchestrator to native numerical functions. Its scalar Phi, Psi and related outputs are catalogued with exact formulas and source context. Their names do not automatically give them the meaning of IIT quantities or experimentally…

System recordSource recovered; static review

github-generator-autoregressive

Generator family · autoregressive

The shared implementation embeds integer-like inputs using a sinusoidal table, combines them with positional weights, applies dense layers and GELU, then tries to produce vocabulary probabilities. Its Train method computes a cross-entropy-like loss without updating weights. The implementation is a…

Generator familySource recovered; static review

github-generator-diffusion

Generator family · diffusion

The shared implementation constructs a linear or cosine noise schedule, adds noise, predicts noise with the local network and performs a reverse-step calculation. Generate begins from random noise. Train compares the predicted and reconstructed noise with mean squared error and returns that loss.

Generator familySource recovered; static review

github-generator-adversarial

Generator family · adversarial

The shared implementation initializes separate generator and discriminator arrays, maps a latent vector through the generator and scores the result with the discriminator. Train computes logarithmic discriminator and generator losses from real and generated inputs.

Generator familySource recovered; static review

github-generator-variational

Generator family · variational

The shared implementation encodes latent means and log variances, samples with reparameterization, decodes the sample and computes a Gaussian KL term. Train combines binary cross-entropy reconstruction loss with a configurable beta multiplier on KL divergence.

Generator familySource recovered; static review

github-generator-flow

Generator family · flow

The shared implementation chains affine coupling transformations and permutations, maps inputs into a latent space, combines a standard-normal log density with accumulated log-Jacobian terms and can generate samples by transforming random latent vectors. Train returns a negative…

Generator familySource recovered; static review

github-generator-energy

Generator family · energy

The shared implementation computes an energy, performs short Langevin-style refinement and emits the refined vector alongside an energy-derived score. Train compares the energy of target data with a short negative-sampling chain. The YINYANG configuration is constructed by the function named…

Generator familySource recovered; static review

github-generator-transformer

Generator family · transformer

The shared implementation reshapes input into a sequence, adds sinusoidal positional embeddings, applies attention and feed-forward operations with residuals and layer normalization, then flattens the output. Train returns mean squared error against the target.

Generator familySource recovered; static review

github-generator-recurrent

Generator family · recurrent

The shared implementation keeps hidden and memory vectors, splits input into steps and selects a basic RNN, LSTM or GRU update. It preserves the resulting state between calls. Train compares the output vector with the target using mean squared error.

Generator familySource recovered; static review

github-generator-statespace

Generator family · statespace

The shared implementation initializes state transition, input and output matrices, discretizes the state equation and processes an input sequence while retaining hidden state. A mamba selector uses the local selective-step path. Train returns output mean squared error.

Generator familySource recovered; static review

github-generator-mixture

Generator family · mixture

The shared implementation initializes a router and expert arrays, selects top-k experts and combines their outputs with routing probabilities. Train returns output mean squared error plus a penalty encouraging the selected probabilities toward a uniform distribution.

Generator familySource recovered; static review

github-generator-config-001

GPT2-Small · local configuration 1

GPT2-Small is local Generator Core registry configuration 1, created by NewGPT2Small in the autoregressive package. Its exact constructor and configuration settings are preserved below.