Let's talk about SearchGPT: Lessons from Building my own Search Engine and Prompt Engineering Tricks
Hi everyone, today I am talking about SearchGPT, a search Engine project that I have been working on. Throughout working on it, I've made a few observations that have improved the chatbot, and I would like to share those with you today. I hope this video is helpful, even for those who want to be better prompt engineers while using chatGPT yourself.
Timestamps:
0:00- Introduction
2:51- Why Search?
5:16- Initial Prompt
7:01- Observation 1: Too many objectives at once
11:23- Observation 2: Knowledge Cutoff and Supplemental Information
14:23- Observation 3: Leverage API & Other (Built-in) Knowledge
18:10- Observation 4: 4000 tokens is "too little" for "everything"
21:12- Observation 5: Context Lifetime and Expiry helps
24:56- Observation 6: "App"ify vs AllGPT
27:28- Observation 7: "Overfitting" Prompts
30:42- Observation 8: Test, test, test (and iterate)
33:05- What's Next
35:03- Conclusion