ChatGPT is a conversational AI system built around GPT, a generative language model that predicts likely next words. Multiple training stages refine this sentence-completion ability into meaningful human conversations. section 1.5
ChatGPT’s capabilities develop through three learning phases:
- Unsupervised learning trains GPT to find language patterns and predict the next word. section 1.5
- Supervised learning refines initial outputs using human-guided examples. section 1.5
- Reinforcement learning further improves conversational responses through trial, feedback, and adjustment. section 1.5
- Applications surrounding the model can add memory, external data, and tools, making interactions appear less stateless and knowledge more expandable. section 1.3
Figure 1.6 The learning stages of ChatGPT
Trace the learning stages that transform GPT’s language prediction into conversational behavior. section 1.5
