🤖

Module 6 - Article 1 - What is GenAI? Generative AI explained

Jul 22, 2025

Overview

This lecture explains generative AI (GenAI): what it is, how it works, its history, current applications, limitations, concerns, and possible future developments.

What is Generative AI?

  • GenAI uses advanced algorithms to create new content (text, images, audio) in response to prompts.
  • It encodes existing data into a vector space and decodes new content by exploring data correlations.
  • Popular GenAI tools include OpenAI ChatGPT, Google Gemini, and Dall-E.

Key Technologies and History

  • GenAI emerged from early chatbots (Eliza, 1960s) and saw major advances with GANs in 2014.
  • Transformers and large language models (LLMs) enabled training on massive text datasets, improving results.
  • Innovations like diffusion models support high-quality image generation and multimodal AI.

How GenAI Works

  • Foundation models (e.g., GPT, Gemini) are trained on large, diverse datasets.
  • GenAI models include encoders (convert input to intermediate vectors) and decoders (generate output).
  • Algorithms include transformers (self-attention), diffusion models, VAEs, GANs, and KANs.

Major GenAI Tools

  • ChatGPT: Multimodal text, image, and voice generation.
  • Google Gemini: Multi-domain AI tools, image, and audio generation.
  • Microsoft Copilot: GenAI embedded in Office, GitHub, and business apps.
  • Perplexity, Anthropic Claude, DeepSeek: Specialized or efficient GenAI systems.

Applications and Business Benefits

  • Automates customer service, document drafting, code generation, and personalized marketing.
  • Increases efficiency, boosts personalization, and improves risk management.
  • Used across industries: finance, legal, manufacturing, and education.

Limitations and Concerns

  • Often lacks source transparency and can generate plausible but incorrect or biased information.
  • Risk of hallucinations, fake citations, copyright/ToS violations, data privacy issues, and high energy use.
  • Raises ethical challenges: bias, deepfakes, data sweatshops, security threats.

Best Practices

  • Clearly label GenAI-generated content.
  • Always vet accuracy, consider biases, and check code/content with other tools.
  • Understand tool strengths/limitations and implement security guardrails.

The Future of Generative AI

  • GenAI adoption is rapid but faces scaling, cost, and accuracy challenges.
  • Research explores agentic AI frameworks, improved trust, and integration into existing tools.
  • Expect ongoing evolution across translation, content creation, 3D modeling, and more industries.

Key Terms & Definitions

  • Generative AI (GenAI) — AI that creates new, original content using learned data patterns.
  • Prompt — User input or query that initiates content generation.
  • Vector space — Mathematical space mapping data based on correlations.
  • Transformer — AI model architecture excelling in NLP and sequence tasks using self-attention.
  • Large Language Model (LLM) — A GenAI specialized in understanding and generating text.
  • GAN (Generative Adversarial Network) — Two-network model generating realistic synthetic data.
  • Diffusion Model — AI model for image generation via transformation between data spaces.
  • VAE (Variational Autoencoder) — Model encoding data into probabilistic latent spaces.
  • Hallucination — AI-generated information that is factually incorrect or made up.

Action Items / Next Steps

  • Review frequently asked questions and key differences between GenAI, LLMs, and traditional AI.
  • Explore recommended readings: generative AI vs. machine learning, GPT-4o vs. GPT-4, and ethical AI frameworks.