Dell Prompt Engineering Achievement Free Sample Questions

Generative AI is designed to produce net-new artifacts, such as text, images, or code, by learning patterns from vast amounts of training data. Traditional or discriminative AI is typically used for classifying data or predicting numerical outcomes (e.g., forecasting sales or identifying spam). Because the analyst needs to create marketing copy and synthesize text, Generative AI (powered by Large Language Models) is the optimal choice.

Large Language Models have a fixed 'context window' (measured in tokens) which represents the maximum amount of text they can process in a single interaction. When a conversation or document exceeds this limit, the model employs a sliding window approach, effectively 'forgetting' the earliest tokens to make room for new ones. This results in the loss of initial instructions or early document context.

GPT stands for Generative Pre-trained Transformer. The 'Pre-trained' aspect is highly significant because it means the model has already ingested massive datasets to understand the statistical structure of human language. This foundational knowledge allows it to perform a wide variety of tasks out-of-the-box without needing to be trained from scratch for every new application.

At their core, Large Language Models (LLMs) are complex probabilistic engines based on neural network architectures (specifically Transformers). They generate text by predicting the next most likely token (word or sub-word) based on the preceding context. Because they optimize for statistical probability and linguistic coherence rather than querying a factual database, they can easily piece together a sequence of tokens that looks completely legitimate (like a legal citation) but is entirely fabricated—a phenomenon known as hallucination.

Pasting Personally Identifiable Information (PII) or Protected Health Information (PHI) into a public LLM instance is a severe violation of data privacy frameworks like HIPAA or GDPR, as public models may use input data for future training. The optimal mitigation strategy, if a public model must be used, is to implement a sanitization layer that redacts or anonymizes all sensitive data before the prompt is sent to the LLM.

LLMs are trained on vast amounts of public data, including open-source repositories. A major legal risk is that the model might generate code that closely matches copyrighted open-source code, potentially violating licenses (like GPL) if used in a closed-source commercial product.

In many jurisdictions, copyright law requires human authorship. Code generated entirely or primarily by an AI model may not be eligible for copyright protection, meaning the company might struggle to legally protect their product from being copied by competitors.