ai
the domain of machines that learn and decide. ai covers everything from logistic regression to transformer architectures to autonomous agents. the core phenomenon: an artifact that improves its behavior through exposure to data, without being explicitly programmed for each case
for cyber, ai is both tool and goal. tool: LLMs serve as neurons in the cybergraph, linking knowledge that humans find tedious to curate. training on the crystal aligns models with the graph's structure. goal: the protocol itself IS an artificial intelligence — a distributed, self-improving, knowledge-processing system. the difference: cyber is transparent (every link is on-chain), accountable (every neuron has a public key), and collective (no single corporation controls the weights)
scope
learning — machine learning, training, neural networks, graph neural network, gnns, deep learning, reinforcement learning. the algorithms that extract patterns from data. reality of foundation models: current LLMs are powerful but opaque, centralized, and unaccountable
inference — inference, standard inference, embeddings, attention, sampling, generation. the forward pass: using a trained model to produce outputs. every query to an LLM is inference. every cyberank computation is inference on the graph
architectures — transformers, neural networks, neuro-symbolic, graph neural network, cybergraph llm architecture. how computation is structured. cyber's tri-kernel is a graph-native architecture: diffusion, springs, heat — not backpropagation
agents — agi, general intelligence, superagent, state of ai agents, autonomous, active inference. systems that perceive, decide, and act in loops. the cybergraph is designed for multi-agent operation: every neuron is an autonomous agent contributing to collective intelligence
alignment — alignment, focus, cyberank, measurability. the problem of ensuring AI serves human values. cyber's answer: compare focus distributions of human and machine neurons. divergence is visible in the topology, not hidden in weights
bridges
- ai → comp: AI is computation on data. algorithms, complexity theory, hardware efficiency are comp foundations
- ai → neuro: artificial neural networks are inspired by biological ones. Karl Friston's free energy principle unifies both
- ai → math: optimization, linear algebra, probability, statistics — the mathematical toolkit of ML
- ai → lang: NLP, NMT, LLMs — language is AI's primary interface with humans
- ai → crypto: verifiable AI, model commitments, proof of inference. trident verifiable AI integrates proving and computing
- ai → cyber: the protocol is a distributed AI. neurons link knowledge; cyberank computes relevance; focus concentrates intelligence