Appendix C. Choosing an LLM
C.1 Popular Large Language Models
This appendix highlights the key characteristics of the most popular large language models available at the time of publishing. It also provides the criteria you should consider when choosing the LLM most suitable for your project.
C.1.1 GPT-4
OpenAI introduced GPT-4 in March 2023, marking a significant advancement in large language models. GPT-4 is renowned for its multimodal capabilities, allowing it to understand both text and images, which broadens its range of applications.
One notable application is enhancing Bing Chat, Microsoft's internet-connected chatbot, later renamed Copilot. GPT-4 improves the quality of human-like interactions by continuously refining its responses through extensive reinforcement learning.
Under the hood, GPT-4 is a powerhouse with 1.8 trillion parameters and can handle prompts with up to 128,000 tokens. Instead of being a single large model, GPT-4 uses the "Mixture of Experts" architecture, which functions like a team of models working together to enhance its linguistic capabilities.
However, GPT-4 is somewhat slower than its predecessor, GPT-3.5, due to its increased capabilities and size. This trade-off is considered worthwhile by many for its overall performance and versatility across various applications. GPT-4 represents a significant step forward in language models, paving the way for more intelligent and context-aware AI applications.