Writing Tailormade Unit Tests with ChatGPT
summary
Shows how to fine-tune an OpenAI model with JSONL prompt/completion pairs so generated tests match an existing testing style, using a Node.js web service with create training and generate endpoints to push data and request unit tests. Demonstrates feeding positive/negative/edge-case tests, warns that training is time-consuming and expensive and that context windows (≈3k–15k tokens) limit chat-based approaches, and suggests WizardLM as an open-source alternative while emphasizing a manual review feedback loop.
AI-written summary · about
chapters (6)
- 0:00Introduction and speaker intro
- 4:23Why unit tests matter
- 9:04Limitations of plain ChatGPT
- 13:00Training custom ChatGPT models
- 18:43Testing and validating the model
- 21:54Challenges and alternatives
related talks
51mTDD with ChatGPT
01h 16mAutomation of the Automation using ChatGPT
21mChatGPT, LLMs, and LangChains - A Beginner's Guide
44mExploring ChatGPT for Improved Observability
47mOperationalization of AI/ML models and use cases for ChatGPT Q&A
36mChatGPT with Laravel & beyond
51mChatGPT as a developer's personal servant
48mTestGPT: Generative AI (AutoGPT, ChatGPT & VisualGPT)