Generative AI
LLM
- Definition
- Use a model that is trained on a large amount of human information to produce similar but new information, guided by text prompts.
How does it work?
Transformer-based AI models are trained to recognize patterns in large amounts of text, and associations between text and images or videos. With this neural-net based information it is possible to tell the model to produce more similar output, based on an input of just text (prompt).
Examples
- LLM, Large Language Model (ChatGPT, Claude)
- Image (Dall-E)
- Video (SORA)
- OpenAI LangChain / LangGraph
When should you use it?
Gen AI is useful in application development, but we're mainly focusing here on the use of Gen AI as part of the application itself. Such an application can
- Act as a customer support chatbot, virtual agent, or shopping assistent
- Do text summarization and content creation for marketing, health care, and news media
- Text translation
- Fraud detection and compliance
- Do text/document retrieval based on a description rather than just text matching (also for private client documents, using RAG (Retrieval Augmented Generation))
- Help a developer write better code and design an application
- Perform simple database queries
- Act like a software agent, using text prompts in stead of function calls
- Multiple Gen AI agents can work together as a multi agent system to perform business tasks
Problems
- A Gen AI agent isn't deterministic: it acts differently next time (although it can be fixed)
- Programming a computer using natural language less precise than using a programming language
- Telling a Gen AI agent what to do is slower than doing it directly, and much more expensive