Large language models are only one slice of the broader AI ecosystem. The platform’s material distinguishes several other families of AI techniques and system architectures that are routinely used in diplomatic‑tech, data‑governance and other policy contexts.
| Category | What it does | Typical examples |
|---|---|---|
| Supervised learning | Algorithms are trained on labelled data (e.g., “cat = 1, dog = 0”). They are used for classification, regression or ranking tasks. | Image‑recognition classifiers, spam‑filtering models [1] |
| Unsupervised learning | Models discover patterns in unlabelled data (clustering, anomaly detection). | Topic‑modeling of diplomatic documents, fraud‑detection systems [1] |
| Reinforcement learning (RL) | An agent learns by interacting with an environment and receiving rewards, useful for sequential decision‑making. | Autonomous negotiation bots, logistics optimisation for peace‑keeping missions [1] |
| Rule‑based / deterministic systems | Explicit logical rules, knowledge‑graphs or symbolic engines that produce exact, auditable results. These are often paired with an LLM front‑end to supply precise calculations or policy‑rule checks. | Symbolic math engines (CAS), policy rule engines, knowledge‑graph query services [2][3] |
| Small / domain‑specific language models (SLMs) | Compact models (millions‑to‑low‑billions of parameters) trained on narrow corpora; they run efficiently on edge devices and are tuned for particular sectors (e.g., maritime law, humanitarian logistics). | Phi‑3, Mistral 7B, open‑source Llama 2 7B [4] |
| Computer‑vision models | Convolutional neural networks (CNNs) and vision‑transformers that process images or video. | Satellite‑imagery analysis for conflict monitoring, facial‑recognition systems for border security [1] |
| Audio / speech models | Speech‑to‑text, text‑to‑speech and voice‑synthesis engines. | Real‑time translation interpreters for multilateral meetings, voice‑activated diplomatic dashboards [1] |
| Generative non‑text models | GANs, diffusion models and auto‑encoders that create images, music or 3‑D assets. | AI‑generated visual art for cultural diplomacy, synthetic data for training security‑analysis tools [1] |
| Hybrid architectures | Combine an LLM front‑end (natural‑language understanding) with deterministic back‑ends (rule‑bases, knowledge graphs, numerical solvers) to deliver both conversational flexibility and mathematically guaranteed outputs. | “Second‑brain” tools used by Singapore’s diplomatic service, where the LLM parses a query and a CAS supplies the exact integral before the answer is rendered in plain language [3] |
These categories illustrate that AI research and deployment span statistical learning, symbolic reasoning, generative media, and lightweight edge models. Diplomatic organisations increasingly draw on a mix of these techniques—choosing the most appropriate type for the problem rather than defaulting to LLMs alone [2][3].