Large Language Models

Updated: 2025-11-24

Large Language Models (LLMs) are language models trained on vast amounts of text data to understand and generate human language. LLMs have one simple objective: to predict the next token given previous context in a sequence. This foundational objective, combined with massive scale, enables them to:

• Text Understanding and Generation: Answer questions, write creative content, translate languages

• Reasoning and Analysis: Break down complex problems, perform multi-step reasoning, make logical inferences

• Information Synthesis: Summarize documents, extract key insights, and aggregate information

• Interactive Assistance: Maintain context in conversations and adapt to user needs

• And more!

Proprietary LLMs

Several commercial models are available through API access with usage-based pricing:

• GPT (OpenAI): Pro, regular, Mini, Nano

• Claude (Anthropic): Opus, Sonnet, Haiku

• Gemini (Google): Pro, Flash, Flash-Lite

Open-Source LLMs

Open-source models can be downloaded and run locally, though they often require significant computational resources (high-end GPUs like NVIDIA H100, H200):

• Llama (Meta): 1B, 3B, 8B, 70B, 405B

• Qwen (Alibaba): 0.5B, 1.5B, 3B, 7B, 14B, 32B, 72B

• Gemma (Google): 2B, 7B, 9B, 27B

The "B" suffix indicates billions of parameters (e.g., "70B" means 70 billion parameters), which roughly correlates with model capability and computational requirements.

Model Tiers

Different tiers of models excel at different tasks:

• Frontier models: Best for complex inference, nuanced understanding, and creative tasks

• Fast models: Optimized for speed and cost-efficiency with good general performance

• Specialized models: Designed for specific capabilities like advanced reasoning or coding

When choosing a model, consider:

• Task complexity: How much reasoning or context understanding is required?

• Latency requirements: How quickly do you need responses?

• Cost constraints: What's your budget for API calls?

• Context window: How much text do you need to process at once?

LLM Access

You can interact with LLMs through multiple interfaces:

1. Web/App Interfaces: ChatGPT, Claude, Gemini

2. API Access: Programmatic access for building applications and integrating into workflows

3. Local Deployment: Running open-source models on your infrastructure

For this course, you will primarily use API access to programmatically interact with LLMs, enabling you to build sophisticated NLP applications.