10 Levels of ChatGPT Prompting: Beginner to Award Winning
Ғылым және технология
Join me on a journey through the 10 levels of prompt engineering techniques. We start with some basic techniques, like basic prompt formatting, then move to more advanced techniques like using personas, the move on to more advanced techniques, like Chain of Thought (CoT).
For my current role, I've needed to squeeze every bit of accuracy, and general helpfulness, I can from GPT-4 and other LLMs. I've spent dozens of hours reading research papers and hundreds of hours prompting large language models - this is my attempt to condense all of that knowledge into 9 minutes.
I hope you learn something from this video. Comment with any questions, and I'll make sure to respond!
***
My custom instructions (system prompt) is available here: gist.github.com/patrickstorm/...
***
0:00 - Intro
0:19 - Level 1 - Basic requests
0:41 - Level 2 - Using formatting
2:10 - Level 3 - Focused requests
3:09 - Level 4 - Give examples
4:00 - Level 5 - Self reflection
4:16 - Level 6 - System prompt / custom instructions
5:02 - Level 7 - Use personas
5:46 - Level 8 - Chain of thought
6:12 - Level 9 - Self prompting
6:47 - Level 10 - CO-STAR framework
Пікірлер: 27
I don’t usually subscribe to new channels, but when I do, it’s because they’re solid. Good job!
@PatrickStormAI
19 күн бұрын
Thanks for the compliment and follow! I’ve got lots more good stuff in the works
@Noqtis
19 күн бұрын
I subbed too, the algo needs more food
this is amazing, it will be practical to sharpen my prompt.
I've actually used all of those prompts coincidentally before, (except the costar) so I subbed just from being happy to see I know what I'm doing lol
@PatrickStormAI
18 күн бұрын
Right on! Co-star is just a good way to organize and not forget everything for a great prompt. So sounds like you know exactly what you’re doing!
Level 5 blew my mind a bit in the simplicity of it. haha.
@PatrickStormAI
19 күн бұрын
It’s almost too easy ha!
That was useful, thank you!
Good luck with the algorithm my dude, got here from my feed. Great video!
@PatrickStormAI
19 күн бұрын
Thanks! My goal is to make good videos, and after maybe 30 of them start worrying about the algorithm 😄
Awesome video, just subscribed
@PatrickStormAI
19 күн бұрын
Thanks so much! More to come, and if you have any specific topics you want me to cover, just let me know
Very informative. I subconsciously use most of these techniques individually in most cases depending on the expected output. Will try implementing COSTAR but it takes time to write such detailed prompts everytime😅😅
@PatrickStormAI
9 күн бұрын
Definitely! I don't use that format all the time, but when I'm asking for complex stuff, I make sure to include all those parts!
number 9 is hilarious.
@PatrickStormAI
18 күн бұрын
Yeah, it’s crazy. The research paper says it outperforms humans like 75% of the time…
Nice video. I knew some but now I have some angles of new inspiration how I wanna tackle some problems I have with ChatGPT. If you have the time and passion I would highly enjoy a video from you about how to get dall-e to stay consistent with an specific art style. It's east to replicate the most known ones but I have a really hard time making it do something that looks drawn but at the same time realistic. Realistic always kinda triggers it to make it close to cgi like even if I specify it by saying realistic proportions +hand drawn. I tried a lot but I don't get a style consistent. Ever it's semi cgi or one of a million different hand drawn styles. Would be glad for some help. I have a huge project in the making I would like a consistent style for my images but I'm kinda stuck :/
@PatrickStormAI
19 күн бұрын
Hey! I do have some experience with that - I briefly worked on a website that used stable diffusion to generate children’s books, it’s been a while though. I’ll add this to my list of videos to make, I’ll definitely need to do some research for it though!
Very insightful. I hadn't considered the impact of being nice, but I make sense of it like this: I assume angry language had largely been removed from training data, and that the future will include more of the same. I assume we curate the data to include positive interactions. Over time AI is trained on language created by nice people - if you say nice things when interactkng, your input will be more firmly inside the training set's "tone". To the social scientists among us - if you can get better results with conversational AI by being nice, and many people work with AI many hours a day. How is this likely to change the way humans interact with other humans?
@PatrickStormAI
16 күн бұрын
Great points! I think you are right. My current read on it is that, on average, intelligent discussions use more polite, civil language - so by using polite language, the language models are more biased toward the intelligent parts of the training data. And I do truly hope you are right that it becomes standard practice to be nice to LLMs and maybe that rubs off on how we interact with people, online and offline!
@JohnBoen
16 күн бұрын
@@PatrickStormAI I have been playing in this space for the last 18 months or so. That is about when I started paying for GPT... I have been doing things 1-9 for a while. COSTAR. That was immediately helpful. What a great template!
I would like to organize the subtitles of this video into Chinese and publish them on my blog. May I have your permission to do so? I will note the source."
@PatrickStormAI
Күн бұрын
Sure! Thanks for asking.
That is not a common knowledge? It is intuitive way to interact with language model
@PatrickStormAI
17 күн бұрын
Some of it definitely is! I made sure to only add in things that are back by research though. I think intuition with language models is built up by playing with them a lot, so good job getting to that place 😊
@cyclejournal9459
2 күн бұрын
@@PatrickStormAI could you link to the study’s you’ve reviewed in this video? 🙏