As a newcomer to this, this was incredibly enlightening! My take-way is that yes, with some proper prompts for complex solutions, you get very close to "I might as well have done the whole thing myself". The point is the scaling: you go through these steps, because you then have a "prompt model" that you can use over and over for similar tasks.
Jonathan, thank you so much for this. I have been at quest where I have been studying many different techniques when it comes to extraction of data in form of products from a database we have. I have now developed a technique, but the hardest part that I have worked with several months is making the llm to update it's chunks from the database as it is a cost/performance balance that we are trying to achieve. We are very close, and we have seen that how we structure our data is as important as the base prompt. We are using retrieval augmented generation also. Can you please comment on this or make a video. I think this will be benifitial to many in the future. Thank you again, very good content.
Your tutorial is superb. If you may, do you have a PDF copy of prompting examples as shown in this video? It would greatly help so we'll just copy and paste to test and actually play with those prompting examples for more interactive learning.
You forgot swearing as a prompt technique. The other day I was prompting a woman with a skirt and kept images of her in a pair of jeans, no matter how many parentheses I put around the word (((skirt))). Finally I lost my patience and wrote "She is wearing a f***ing skirt!" and it worked like a charm :D
@@LatentAI Cool! Yes, I totally think such a stream would be nice. Also a collection / dataset of high quality prompts would be nice. Let me know if you know of any. I am thinking of building an app around this to make it fun to learn.
just curious, why use chinese word "host" as your channel's logo, and "shujin" sounds a lot like the chinese pronunciation, or the japanese pronunciation?