Samantha Stark is the founder and chief strategist of Phyusion.
Basic AI concepts
Artificial Intelligence (AI) – A wide range of computer science dedicated to building machines that can perform tasks that normally require human intelligence, such as recognition, reasoning, learning, and language use.
Agent Function – The ability of AI systems to autonomously pursue goals: to make decisions, planning, and action with minimal human intervention, while respecting predefined constraints.
Generation AI – A model that allows you to learn patterns of data and create new content (text, image, audio, or video) similar to the training distribution given a prompt.
aeo (responds to engine optimization) – Practices to optimize content and prompts to improve visibility, relevance and performance within generation AI systems such as CHATGPT, Gemini, Claude.
Major Language Models (LLM) – Large-scale neural networks (GPT-4, Claude 3, Gemini 5) trained in huge text corpus to understand and generate human-like languages.
Trans Architecture – The deep learning model introduced in 2017 uses auto-joints to process the entire sequence in parallel, forming the technical backbone of modern language, vision and multimodal models.
Machine Learning (ML) – A subfield in AI that allows computers to learn patterns from data and improve them in tasks without being explicitly programmed with task-specific rules.
Natural Language Processing (NLP) – AI branches focused on enabling computers to understand, generate and interact with human language.
Audio cloning – Analyze the vocal characteristics of speakers and enable techniques to integrate speeches that mimic audio, personalize or consistent audio content.
Contextual perception – LLM's ability to incorporate conversation history, user-specific data, or retrieved documents ground the output in the relevant context.
Multimodal Generation – Creating or interpreting content across multiple modalities (combination of text, images, audio, or video in a single model interaction).
Microsoft Copilot -Specific Terminology
Microsoft 365 Copilot – AI assistants built into Microsoft 365 apps (Word, Outlook, PowerPoint, Excel, Teams, etc.) help you draft content, summarise information, analyze data, and automate workflows.
Enterprise-grade security – Microsoft 365 Copilot inherits platform compliance, privacy, and security controls (cryptography, identity, data residency, and administrative governance) to protect your corporate data and IP.
bizchat – Co-pilot features using natural language queries to extract insights from Autsook, Teams, Onedrive, SharePoint using natural language queries – prevent meetings, emails, and documents from useful updates.
Common AI tools
chatgpt – Openai's conversation interface with GPT model that can answer questions, generate content, and assist in a variety of tasks.
Custom GPT – A tailored version of OpenAI's GPT model, which consists of specific instructions, unique knowledge, and tool integrations for organizations or individuals to resolve domain-specific problems.
Claude -Anthropic's AI assistants have noted about constitutional adjustment approaches, large context windows, and subtle reasoning.
Claude Project – Claude's workspace feature allows users to upload documents and collaborate with the model across iterative sessions that are different from custom GPT.
Gemini – Google Deepmind's multimodal AI model (formerly Bird) is available on Google Workspace and other Google products.
Dall-E – Image model from Openai text that generates the original image from natural language description.
Mid Journey – Independent generated image tool widely used to produce high-decision artwork from text prompts.
Sora – Openai text-to-video model that generates short, realistic and imaginative video clips from text prompts.
Hagen – A platform for creating AI-creating videos with customizable digital avatars and narration.
evelenlabs – Advanced AI voice technology for natural sound speech synthesis and speaker-specific voice cloning.
Confused – Answer AI search and question engine that combines large search with LLMS to provide concise answers cited.
runwayml – Generic AI tools for creating and editing videos, including text-to-video generation, video-to-video generation, image animation, and more.
Fundamentals of rapid engineering
prompt – Text (or multimodal) input provided to the AI model is provided to elicit the desired response.
Fast engineering – Crafting of prompts to guide AI systems to high quality, task-related output.
Prompt Template – Reusable, rapid structure that encapsulates best practices and quickly adapts to a variety of tasks.
temperature – Generation parameters that control randomness (0-1 or 0-2 depending on the model). Higher values give you more diverse and creative output, while lower values give you more critical results.
Repeated refinement – A workflow in which the user reviews the AI output, provides feedback or clarification instructions, and repeats the process until the response meets quality standards.
AI Limitations and Considerations
Hallucinations – If the AI model actually produces output that is incorrectly manufactured, or non-sensic yet sounds plausible.
Training Data Cutoff – The most recent date of data used to train the model. Events that occur after this date are unknown to the base model unless supplemented by external search.
Token limit – The maximum number of tokens (almost word or word piece) that can be processed in a single prompt and response cycle that constrains the length of the context.
bias – Systematic inaccuracy or inequality in AI output due to imbalance, representative, or biased training data.
Ethical AI Use – Principles and practices that promote the responsible development and deployment of AI, including transparency, accountability, fairness, privacy protection, and mitigation of harmful stereotypes.
Key concepts of corporate PR leaders
GuardRails / Policy Enforcement – Technical and procedural controls that limit AI output to brand, legal, and compliance standards prevent the release of permits or unbranded content.
Loop human (hitl) – A governance workflow in which human reviewers approve or modify AI-generated content prior to publication, ensuring accountability and quality control.
Emotional analysis – AI techniques to detect and classify emotional tones in text or voice, allowing real-time monitoring of viewer responses and message impact.
This article is a preview of the content available to members of Ragan's AI Strategy Center.
