AI 101 - LLM

Chapter 1: Introduction

• Yellow Lens: Introduction to the discussion about large language models (LLMs) and retrieval augmented generation (RAG).

Large Language Models (LLMs):
  - LLMs are AI systems that generate human-like text based on substantial training data.
  - The training involves scraping trillions of pieces of information from the Internet, articles, books, social media, and more, creating a database that enables the model to make predictions and produce language.
  - LLMs function somewhat like robots that provide responses based on their programming and training and are limited to the information they have processed.

1. Retrieval Augmented Generation (RAG):
     - Describes a system that moves beyond LLMs by enhancing the generated text with real-time contextual information.
     - Definition:
       - Retrieval: The system retrieves relevant information from external sources.
       - Augmented: Enhancing or providing more depth to the information retrieved from LLMs.

2. Comparison of LLMs and RAG:
     - An LLM can generate a summer program for the youth but lacks local context.
     - RAG goes further by analyzing specific community demographics and preferences before generating a tailored program.

3. Agents:
     - Refers to AI systems that execute tasks autonomously (e.g., Gamma, Mixo, Otter).
     - Agentic AI:
       - Functions without external input and operates continuously without needing constant user intervention.

Understanding Modern AI:
  - By understanding LLMs, RAG, and agents, one grasps the essentials of modern AI technology.

Chapter 2: Copilot And Perplexity

• Development of LLMs:
    - Early LLMs used conversational data, notably from Reddit, to train algorithms in a real and engaging manner.
    - These models learn from vast amounts of data to understand language patterns and facilitate predictions.

• LLM 'Brains':
    - The term 'brains' refers to powerful standalone LLM systems capable of performing multiple complex tasks.
    - Examples include ChatGPT, Claude (Clawd), Google Gemini, Grok, DeepSea.

• AI-Powered Tools:
    - Tools such as Microsoft Copilot and Perplexity utilize underlying LLM technology to deliver specific functionalities.
    - Microsoft Copilot is built on ChatGPT, while Perplexity integrates various LLMs along with live internet searches.

Chapter 3: ChatGPT

• OpenAI:
    - OpenAI is the company responsible for developing ChatGPT. It is essential to differentiate the company's name from the product.

• ChatGPT Characteristics:
    - Nicknamed as a "smart, witty conversationalist," suitable for generating creative ideas and conversations.
    - Capable of assisting in writing, strategy, coding, teaching, and reasoning.
    - Great for brainstorming, strategy and problem-solving, and quick learning.

• Functionality:
    - Functions using the Paraphrased Definition of "GPT":
      - G: Generative (capable of generating responses).
      - P: Pre-trained (developed from existing data).
      - T: Transformer (transforms input into intelligent output).

Chapter 4: Need Any Type

• Microsoft Copilot:
    - Functions like a supportive co-worker, taking on practical tasks across Microsoft Office (Word, Excel, PowerPoint).
    - Not designed for lighthearted interaction; focused on efficient task completion as an intelligent assistant.
    - Ideal for tasks requiring detailed analysis and summarization.

• Claude:
    - Developed by Anthropic, Claude symbolizes wisdom and ethics in AI development.
    - Designed with ethical values to handle nuanced and sensitive topics, offering wise and careful advice.
    - Particularly effective for summarizing lengthy documents and computer code.

• Google Gemini:
    - Known as the "multimedia genius," best for tasks requiring multi-format outputs, such as videos and visually rich presentations.
    - Integrates seamlessly with Google’s capabilities in research and multimedia generation.

Chapter 5: Social Media Whisperer

• Grok:
    - Acts as a direct link to social media (formerly Twitter), providing real-time insights and cultural commentary.
    - Useful for tapping into current events and trends that populate social media platforms.

• Perplexity:
    - Distinguished for providing accurate academic-level references, citing sources appropriately in formats like APA and MLA.
    - Specifically designed for students and researchers to provide real, reliable links to information drawn from various LLMs and academic sources.

• DeepSea:
    - Represented as the “no-frills problem solver,” characterized by efficiency in technical problem-solving, particularly in coding, math, and engineering without personality or excessive chatty features.
    - Not an American product, it utilizes OpenAI’s technology controversially and raises concerns about data privacy given its Chinese ownership.

Chapter 6: Conclusion

• Summary and Caution:
    - Users should be aware of the political and ethical implications involved in using AI technologies, particularly those developed outside the United States, such as DeepSea, due to privacy issues and data ownership rights.
    - There are potential risks associated with the technology's affiliations and operations under a non-democratic regime.
    - The importance of understanding the genre and functionalities of each AI tool can significantly benefit users' interactions with AI.

• Final takeaway:
    - By familiarizing oneself with the different AI systems, their functions, and underlying ethical concerns, users can gain a distinct advantage over the general populace in understanding AI technologies.