Building Generative AI Applications
Materials — open to everyone, no sign-in
Topic: Building Generative AI Applications
Presenter: denny wang
Additional Resources:
Building Generative AI Applications
Coach Ken LinkedIn:
https://commitway.com/linkedin
WeChat QRCodes
| | | 职场提升俱乐部 |
Fine-tuning is a process in machine learning where a pre-trained model is further trained on a specific dataset or task to improve its performance for that context. It’s like tailoring a general-purpose model to excel in a specific application by providing it with additional, focused training data.
Prompting refers to the process of providing specific input or instructions (called a “prompt”) to guide a pre-trained model, like ChatGPT, toward generating desired outputs. Instead of altering the model’s parameters (as in fine-tuning), prompting focuses on crafting effective input to elicit high-quality and relevant responses.
RAG stands for Retrieval-Augmented Generation, a hybrid approach in natural language processing (NLP) that combines retrieval of relevant information with generative models to produce accurate and contextually informed responses. This technique enhances the ability of generative models to answer questions or perform tasks by grounding their output in real-time, external knowledge.
Fine tuning leads to a couple challenges:
It’s hard to fine tune when the baseline model has a new version
Fine tuning may lead to destroy some ability in the baseline model. It’s hard to detect if the baseline model has been negatively impacted after fine tuning.
Agent can be developed with code or framework. Framework provides convenience but limits flexibility
Emerging areas:
Distributed training
Multi-model orchestration
no-code/low-code AI platforms. Provide data. It will generate a model
autoML
Model interpretability and tracing
Industry-specific AI systems
Customer support as a platform
Automated workflow for business.
Most professional can automate work through automation framework
SDE: highly complex system requiring integration
GenAI:
AI coding
low/no-code
Architecture
SDE needs to improve:
System design
Code review
Business understanding
Cooperate with AI
Some skills are deprecated:
Basic coding / writing
LangChain
Coordination
Memory
Plans
Common memory types
Conversational buffer
Vector DB
Common types of chains
Sequential
Multi-prompt
Router chain
How to prompt AI
Ask positive and negative questions
E.g. what are the alternatives? What is the disadvantages