• Ref: https://www.udemy.com/course/aws-ai-practitioner-certified/learn/lecture/44887829
• Contents:

Capabilities and Challenges of GenAI - Overview

• Capabilities of Generative AI
  • Adaptability
  • Responsiveness
  • Simplicity
  • Creativity and exploration
  • Data efficiency
  • Personalization
  • Scalability
• Challenges of Generative AI
  • Regulatory violations
  • Social risks
  • Data security and privacy concerns
  • Toxicity
  • Hallucinations
  • Interpretability
  • Non-determinism
  • Plagiarism and cheating

Toxicity

• 🔧 Generating content that is offensive, disturbing, or inappropriate
• 💡 Defining what constitutes “toxicity” can be a challenge
• Boundary between restricting toxic content and censorship
• What about quotations of someone that can be considered toxic? Should they be included?
• Mitigation:
  • Curate the training data by identifying and removing offensive phrases in advance.
  • Use guardrail models to detect and filter out unwanted content.
• Diagram

Hallucinations

• 🔧 Assertions or claims that sound true, but are incorrect
• This is due to the next-word probability sampling employed by LLM
• This can lead to content that may not exist, even though the content may seem plausible