Track Title

Artist Name

Track

AI is transforming industries and revolutionizing how we work, interact, and innovate on Salesforce. To fully harness its potential and get the most out of this course, you first need to speak the language of AI – understanding the terminology and concepts of this exciting field. Without further ado, let’s explore the key terms that you’ll need to know for your AI journey!

The AI Fundamentals Podcast

Episode 2: Core Terminology

Artificial Intelligence (AI): The umbrella term for machines or software that simulate human intelligence.

Machine Learning: A subset of AI where models learn from data to make predictions without being explicitly programmed. It’s like teaching a computer by showing it lots of examples, and over time, it gets better at making predictions or decisions based on that experience, just like how humans learn from practice.

Neural Networks: A computer system inspired by the brain, made of connected “neurons” that process data. By learning from examples, it recognizes patterns and improves tasks like image recognition, speech understanding, and predictions.

Deep Learning: A subset of machine learning that uses neural networks with many layers to process complex data like images, text, and audio.

Generative AI: A type of AI that creates new content (e.g., text, images) based on the data it was trained on.

Natural Language Processing (NLP): AI that processes and understands human language. It’s the backbone of virtual assistants like chatbots.

LLM (Large Language Models): Advanced machine learning models that generate human-like text, answering questions and summarizing information.

Generative Pre-trained Transformer (GPT): A type of AI model that uses deep learning and NLP to generate human-like text based on the input it receives. It is pre-trained on large datasets of text to learn the structure of language, enabling it to understand and generate coherent, contextually relevant responses.

Prompt: Refers to the input or query given to an AI model to generate a response or perform a specific task. This input can be a sentence, question, or statement that guides the model on how to respond or what kind of information to retrieve or generate.

Prompt Engineering: The process of designing and refining prompts to elicit the desired output from an AI model, particularly in natural language processing (NLP) systems like GPT models. 

Structured Data: refers to data that is organized and formatted in a way that makes it easily searchable and usable by both humans and machines. It is typically arranged in predefined formats, such as rows and columns in a database or a spreadsheet, with clearly defined relationships between data points. Each piece of data fits into a specific field, such as name, date, or amount, allowing it to be efficiently stored, queried, and analyzed. (Common examples include Excel spreadsheets and CSV files)

Unstructured Data: Refers to information that does not have a predefined organization or structure, making it difficult to store, search, and analyze using traditional methods like relational databases. Unlike structured data, unstructured data doesn’t fit neatly into rows and columns, and it often includes a wide variety of formats that can contain text, multimedia, or other types of content. (Common examples include Emails, text documents and social media content)

Grounding: Infuses LLM prompts with your internal Salesforce data to “ground” the prompt with relevant context.

Hallucination: Instances where the AI generates information or responses that are not accurate, factual, or grounded in the data it was trained on. These “hallucinations” often appear coherent and convincing, but the content may be incorrect, fabricated, or irrelevant. Reducing hallucinations is a key challenge in AI development, and efforts to improve accuracy involve refining model training, improving prompt engineering, and introducing verification mechanisms to ensure the AI’s outputs are based on reliable sources or data.

Red-Teaming: The process of rigorously testing, challenging, and probing an AI system to identify its vulnerabilities, weaknesses, or potential risks. The goal of red-teaming is to simulate adversarial conditions or extreme scenarios in order to assess how the AI behaves, whether it makes errors, produces harmful outputs, or is susceptible to exploitation.

Zero Data retention: A policy or approach where an AI system does not store, retain, or keep any data provided by users after it has processed and responded to a request. This means that once the AI has completed its task, all input data (such as queries, documents, or personal information) is immediately discarded and is not stored for future use, analysis, or learning.

Toxicity: Refers to harmful, offensive, or inappropriate content generated by AI systems, often involving language that is abusive, hateful, or discriminatory. Toxicity detection is an important task in natural language processing (NLP) and AI ethics, where models are trained to identify and filter out such harmful content. Reducing toxicity in AI-generated responses is essential for creating safe and inclusive environments, especially in public platforms and social media.

Parsing: The process of analyzing and breaking down input data, typically language or text, into structured components to understand its meaning. Parsing is fundamental in tasks like speech recognition, language translation, and text understanding. It helps AI systems interpret complex input data for decision-making or further processing.

Now Drop In To Focus

Now you know how to speak the language of AI. Test your knowledge with the quiz! 🚀

Quiz Time!