A.I Guides
AI Field Guide
A growing visual reference for understanding AI and computing. Start with a concept, connect it to an analogy, then explore how it works.
Topic previews are ready to explore. Full explanations and illustrated sheets will be added after review.
4 topics
Hardware & capacity
What runs the workload, and what lets it fit.
GPUsProcessing power
GPUs perform many suitable calculations in parallel.
Picture it
Think of processing throughput separately from the amount of memory available.
In this topic
Parallel work; CPU and GPU roles; workload bottlenecks.
Illustration & full explanation pending review.
VRAMMemory beside the GPU
VRAM holds working data close to a discrete GPU.
Picture it
The truck bed holds the load; its capacity is different from how quickly the truck moves.
In this topic
Capacity versus bandwidth; model working sets; offloading.
Illustration & full explanation pending review.
RAMThe working space
System RAM holds data that programs are actively using.
Picture it
A workbench keeps current materials close at hand; storage is the filing cabinet.
In this topic
Working data; capacity versus speed; RAM and VRAM.
Illustration & full explanation pending review.
StorageFiles between sessions
Storage keeps files and model weights between sessions.
Picture it
The warehouse holds supplies before they are brought to the workbench.
In this topic
SSD and HDD; loading models; storage versus working memory.
Illustration & full explanation pending review.
4 topics
Models & knowledge
Learned capability, current context and retrieved information.
LLMsLanguage models
An LLM is a trained language model that processes input and generates responses.
Picture it
The trained operator is a useful analogy for learned capability, not a claim that the model is a human mind.
In this topic
Models versus applications; generated answers; errors and limitations.
Illustration & full explanation pending review.
Models & parametersLearned settings
Parameters are learned numerical values within a model.
Picture it
The knobs on a soundboard change how it behaves; their number alone does not establish the quality of the result.
In this topic
Parameter scale; total versus active parameters; model evaluation.
Illustration & full explanation pending review.
ContextInformation in view
Context is the information available to a model for the current response.
Picture it
A desk can hold only so much briefing material at once.
In this topic
Tokens; context windows; current context versus saved memory.
Illustration & full explanation pending review.
RAGRetrieving relevant material
Retrieval-augmented generation supplies relevant source material to help a model answer.
Picture it
A runner fetches useful pages from the archive instead of placing the entire library on the desk.
In this topic
Retrieval; embeddings; source quality; grounded answers.
Illustration & full explanation pending review.
4 topics
Connections & interfaces
How software requests capabilities and exchanges information.
APIsRequesting a capability
An API defines how software can interact with another component.
Picture it
A request form specifies what can be asked for and which information it needs.
In this topic
Interfaces versus calls; request and response; authorization.
Illustration & full explanation pending review.
MCPA shared protocol
Model Context Protocol standardizes how AI applications connect to tools and context.
Picture it
The menu, order format and returned result all matter; MCP is more than listing a menu.
In this topic
Tools, resources and prompts; clients and servers; discovery and execution.
Illustration & full explanation pending review.
SDKsA development toolkit
A software development kit provides tools for building with a system.
Picture it
The starter kit can include libraries, examples and documentation.
In this topic
SDK versus API; reusable libraries; developer tools.
Illustration & full explanation pending review.
Local & cloud AIWhere the model runs
Local models run on your equipment; cloud models run on a provider's equipment.
Picture it
Use capability at your own base, or call an outside service.
In this topic
Local software versus local models; connected tools; offline limits.
Illustration & full explanation pending review.
4 topics
Agents & workflows
How actions, coordination and oversight fit together.
AgentsModels with tools
An agent system combines a model with software that can use tools and carry out steps.
Picture it
The operator needs equipment and access as well as instructions.
In this topic
Model and runtime roles; tool use; task state.
Illustration & full explanation pending review.
AutomationRunning a routine
Automation runs specified actions without a person issuing each step manually.
Picture it
An alarm triggers a drill; the routine can include planned branches.
In this topic
Triggers; conditions; repeatable workflows.
Illustration & full explanation pending review.
OrchestrationCoordinating the work
Orchestration manages how tasks, tools and workers fit together.
Picture it
The coordinating role assigns work, waits for results and decides what follows.
In this topic
Dependencies; handoffs; fixed rules and model-led decisions.
Illustration & full explanation pending review.
PermissionsWho may do what
Permissions define which resources and actions a system allows.
Picture it
Clearance controls access; a job title or impressive model does not grant it.
In this topic
Boundaries; approvals; oversight and controls.
Illustration & full explanation pending review.
Based on the AI & Computing Field Manual. Analogies are learning aids; each topic will include its technical limits.