Artificial Intelligence Overview

Artificial Intelligence (AI)

• AI is a computer science branch where systems use data to infer, task, and solve problems using human-like reasoning.

Bias

• Bias is a systematic error in computer systems.
• It produces unfair outcomes, differing from intended function.
• Bias arises due to inaccurate assumptions during machine learning.

Corpus

• A corpus is a substantial collection of textual data used for training Large Language Models (LLMs).

Domain Adaptation

• Domain adaptation integrates organization-specific knowledge.
• It modifies the prompt and foundation model.

Fine-tuning

• Fine-tuning adapts a pre-trained language model.
• It trains the model on a specific, smaller dataset related to the task.

Generative AI Gateway

• The gateway provides normalized APIs for interacting with foundation models and services from various vendors.

Generative Pre-trained Transformer (GPT)

• GPT is a family of language models.
• They're trained on large text datasets to generate human-like text.

Grounding

• Grounding adds context to the model by integrating domain knowledge and customer information into the prompt.

Hallucination

• Hallucination occurs when the model outputs semantically correct text but is factually wrong or nonsensical.

Human in the Loop (HITL)

• HITL models require human interaction during the process.

Hyperparameter

• Hyperparameters control the training process.
• They exist independently of the model's generated structure.

Inference

• Inference is the process of generating content from a model.

Inference Pipelines

• Inference pipelines are sequences of steps to complete generation tasks.
• They involve prompt processing, model interaction, result moderation, and delivery.

Intent

• Intent represents a user's goal in interacting with an AI assistant.

Large Language Model (LLM)

• An LLM is a large neural network trained on substantial text data.

Machine Learning

• Machine learning focuses on computer systems that learn, adapt, and improve using data feedback.

Model Cards

• Model cards provide detailed performance information on models.
• Information includes inputs, outputs, training, optimal use conditions, and ethical considerations.

Natural Language Processing (NLP)

• NLP uses machine learning to process human language.
• LLMs are one NLP approach.

Parameter Size

• Parameter size refers to the number of parameters in a model for processing and generating data.

Prompt

• A prompt is a natural language description of a task.
• It acts as input for an LLM.

Prompt Chaining

• Prompt chaining breaks complex tasks into smaller steps.
• It connects the steps for a more specific and improved result.

Prompt Design

• Prompt design involves optimizing prompts to improve model response quality and accuracy.
• It involves understanding and adjusting prompt structure for better results.

Prompt Engineering

• Prompt engineering is the scientific process of improving model performance and reliability through systematic prompt structuring.

Prompt Injection

• Prompt injection is an approach used to manipulate or control the model's output.
• It's a method to perform actions not intended for the model.

Prompt Instructions

• Prompt instructions are natural language instructions integrated into a prompt template.
• Instructions are part of the prompt sent to the LLM.

Prompt Management

• Prompt management provides tools for building, organizing, managing, and distributing prompts.

Prompt Template

• A prompt template is a string with placeholders.
• The placeholders are substituted with data values for the final prompt.

Retrieval-Augmented Generation (RAG)

• RAG uses knowledge bases or information retrieval to provide relevant context for prompts.

Semantic Retrieval

• Semantic retrieval uses similar historical data for better model accuracy.

System Cards

• System cards are an extended version of model cards.
• They cover the entire system's operation (components, models, processes).

Temperature

• Temperature controls the predictability and variety of model output.
• High temperature = diverse responses, Low temperature = consistent responses

Toxicity

• Toxicity refers to various forms of inappropriate, offensive, harmful, or abusive language.

Trusted AI

• Trusted AI is a set of Salesforce guidelines for responsible AI development and implementation.

This quiz explores fundamental concepts of artificial intelligence, including bias, corpus, domain adaptation, fine-tuning, and generative models. It is designed to test your understanding of how these elements work together in the field of AI and machine learning.