How a chatbot like ChatGPT or Claude is made
Process map by Brian. 25 steps, 2 layers. Tags: ai, tech, science.
- The AI lab: Collect text, code and images from the web and licensed sources
- The AI lab: Clean and filter the data
- A giant computer cluster: Pretraining: predict the next word, trillions of times
- The AI lab: Fine-tune on examples of good answers
- Testers and red teams: Humans rank the answers and the model learns what people prefer
- Testers and red teams: Safety testing: try to make it do harm
- The AI lab: Does it pass the safety and quality tests?
- Users: Release it to users
- Users: People use it, give thumbs up and down, and find new failures
- The AI lab: Update and release the next version
- The AI lab: Fix, retrain and retest