Improve ChatGPT Prompts with Priming (Part 2)
HTML-код
- Опубликовано: 24 апр 2023
- The second part of this series describes Level 5 Priming!
Everyone is pitching the best ChatGPT Prompt... Try GPT Priming instead:
Priming is the best way to formulate better results by inputting related information before requesting a response:
-Allows for variability in your GPT chat dialogue later in the string
-Eliminates the need to create multiple chats
-Produces more precise results
-Improves your overall experience
ChatGPT Resources:
Not another newsletter!! You've got plenty of those... This email won't discuss the news and it will only show up when I've got something very worthwhile to share like:
- Incredible new ChatGPT Prompts,
- Excellent marketing tools,
- Powerful automated workflows
- The Most useful marketing tips and tricks you can imagine
lawtonsolutions.com/subscribe/
You can opt-out anytime with one click and it's totally free.
Free resources for Role Options, Tone/Writing Style Options and format
lawtonsolutions.com/resources/
Other related AI Priming Videos:
lawtonsolutions.com/How-To-AI/
LEVEL 5 PRIMING
FIRST:
I am going to provide you with several variables that I will define via the following = [variable]. I want you to remember these variables for later use. Then we will use these variables to create specific content that I request. Please confirm you understand these instructions by responding, “Read,” and we will proceed to the next step.
THEN:
I want you to store the following example of X in the variable
[Variable 01].
[Variable 01] = "populate this with the example information"
THEN ADD MORE:
I want you to store the following example of X in the variable
[Variable 02].
[Variable 02] = "populate this with the example information"
FINALLY, HAVE CHATGPT EXECUTE THE TASK BASED ON THE VARIABLES:
Now that you have ingested these different variables, please act as an expert "XYZ" with two decade of experience. Based on the variables, [Variable 01] and [Variable 02], please write "CONTENT" with the following components:
Being in IT for decades, I never thought of using variables in CHATGPT or GPT4 this way. This is awesome and really appreciate in sharing this valuable content!!!
Thank you. Great to hear it’s been helpful. Much appreciated.
I have not enough words to say how grateful I am. This has been practically a full course, and so inspiring!
Thank you so much. Glad to hear it was helpful
Great instructional videos on how to get the best from ChatGPT. Will definitely try this formula of analysis before generation and see where it takes me. Thank you!
Thank you and good luck! Please let me know if you have questions.
Loved the priming videos. Ive been working on the prompting and this makes sense.
Would love to see the giraffe joke and get your audience opinion on how good it is!
Thanks for the kind words. Unfortunately, I deleted that conversation after I made the video. I have way too many chats to sort through each time I want to reference prior Chat dialogues. Actually, I'm working on a new video that discusses a way to solve that problem in ChatGPT. Stay Tuned!
Dude…this is GOLD ! Thanks for sharing
Glad you enjoyed it! Much appreciated.
Your content is next-level. Subbed.
Thank you. Glad to hear you’re enjoying it. Much appreciated!
I was actually struggling with going further into my prompting because of the struggle of having to create multiple chats to test outputs and prompts
Same challenge for me as well but this seems to be working.
Glad it's working for you. Perhaps some day it will be able to consume more information in one shot. Until then...
Wow! Me too but now it’s working!
Great idea
Thank you. Best of luck. Please let me know if you have questions.
Hey @LawtonSolutions, I ran into your two videos about priming just now. It's really great, thanks. I was just thinking of using this method for creating instructions / meta prompts. Even though GPT4 can now handle 128k tokens, I was wondering whether you could move the in depth description of each comedians (if I use your example) into separate content files and make them available via RAG. The same, of course, could apply to your tables which you could also "off load". Have you ever tried this?
Thank you. I have tried this using different content scenarios (not comedians). It works with some success but the prompts need to be fairly mature and refined. Also the content files provided for each scenario needs to be structured similarly to work best. Yes, it works but not as well as I’d like (yet).
Great video. Any issues with maxing out tokens by having a number of variables 'loaded'?
Thank you! Good question… the quick answer is yes, there’s a limit. In this example, the content absorption and analysis isn’t critical. If Chat doesn’t fully consume the persona or style of the comedian, the end result isn’t too mangled. However, if you’re looking to maintain quality for a bigger data input, I’d recommend having Anthropic’s Claude consume the data and provide a summary. Then, input that summary into ChatGPT as a variable.
I hope that helps!
The good news is OpenAI continues to expand the token size with new versions, so it may soon be competitive with Claude.
So, I gave chat gpt the instructions that you mentioned in the first prompt. It is refusing to follow the instructions! After 10 minutes I had to finally yell (writing in al caps) to just resond with the word read as it is only a one word response. Chat gpt responded with the following "I can definitely help you remember those variables and use them for creating content later on. So, "Read," let's proceed!"
That’s only happened to me a couple of times. However, when it does happen, I simply copy my prompt and start a new chat. Typically it doesn’t happen in the new prompt. Let me know if you see it again.
Have you tried using other elements of programming to prime ChatGPT aside from variables such as IF/THEN statements?
Thanks for the wonderful question. Yes, I've used other elements of programming with mixed success. While it can certainly write code (and even check itself), executing code is a very different story. Unfortunately, I haven't found it to have the power or reliability of typical programming (though I'm not much of a programmer). I view ChatGPT as a front end, brainstorming tool. I ask it to provide me with the "world of the possible." Then I separately pursue the alternatives and return to Chat with specific questions. Perhaps someday it will get there.
I hope that helps... Though I fear not enough.
I am inspired by this vid, please allow me to use this video as reference for my notes in prompting guide
Thank you for the kind words. Please simply reference the video and the source in your prompting guide.
Are variables a chatgpt plus feature? When I type in your prompt, I get "I apologize for any confusion, but as an AI language model, I don't have the capability to remember variables or store information across different interactions. Each conversation with me is treated as a separate session, and I don't retain any information from previous interactions. I can assist you with generating content or answering questions based on the current conversation, but I don't have a persistent memory of variables or past interactions. If you have any specific requests or questions, feel free to ask, and I'll do my best to assist you."
Thanks for your comment. It's a great question. I've used this prompt structure successfully on both the free and plus versions. I used it successfully on three different accounts this morning so I'm a bit baffled. I'm assuming you're starting with the first part:
"I am going to provide you with several variables that I will define via the following = [variable]. I want you to remember these variables for later use. Then we will use these variables to create specific content that I request. Please confirm you understand these instructions by responding, “Read,” and we will proceed to the next step."
....Then you're getting that response from chat?
Based on the response you shared, Chat seems to think you're trying to use information across chat sessions, which won't work of course. Can you please share the prompt you used to start (removing any private info)?
@@LawtonSolutions My prompt is exactly what is in your video: "I am going to provide you with several variables that I will define via the following = [variable]. I want you to remember these variables for later use. Then we will use these variables to create specific content that I request. Please confirm you understand these instructions by responding "Read" and we will proceed to the next step."
I wonder if it is a GPT-3.5 vs. GPT-4 thing? I tried your prompt on Bing Chat and it accepted it more like your video. My understanding is that Bing uses GPT-4, but my free version of ChatGPT is using GPT-3.5.
@@semosancus5506 Darn it. I'm sorry to hear this. I've asked several friends with free accounts to try it on GPT-3.5. They are all having success. Can you please try giving GPT-3.5's response a thumbs down? Then provide it with feedback that you're not asking it to use variable across different interactions. Then run it again? Alternatively, when you get that response, simply ask GPT-3.5 to check itself, indicating the you're not asking for variables across different interactions and that it is mistaken. Chat is surprisingly good at correcting itself. Please let me know how that works.
@@LawtonSolutions So that did work. Telling ChatGPT that I wanted to use the variables within the conversation helped it understand. So I started a new chat and changed the initial prompt to add the clarification that the variables will be used in this conversation and that also worked.
"I am going to provide you with several variables that I will define via the following = [variable]. I want you to remember these variables for later use within this conversation. We will use these variables to create specific content that I request. Please confirm you understand these instructions by responding "Read" and we will proceed to the next step."
That’s such great news! Glad to hear it!
I understand ChatGPT is only able to reference the last 3000 words of a convo. Does this mean that if the variable set up prompts are beyond the last 3000 words, future mentions of the variable would be meaningless to it?
Things have changed quite a bit now that ChatGPT has been updated. The input limit has changed significantly so the amount of content it can reference per dialogue box has likely increased significantly as well. I will continue to test how much it can meaningfully absorb but this may mean we can drop the entire prompt into one dialogue box instead of using multiple steps. Let’s hope so!
Am I the only one who thought this was Jason Kidd
Ha! Compliment? Honestly, I’m not sure. If you met me in real life you’d say Stanley Tucci.
@@LawtonSolutions I love your video n I’ma have to look up Stanley! Thank you for yo powerful intel on gpt tech!
Shem! Thank you so much!! I really appreciate it!!
OK I'm Clueless, but I don't understand what you are trying to do here. When you say store example of x, is x an example of comedy (in this prompt you are building)? And why multiple variables? I literally have no clue what is going on here. Can you explain it to a child? Thanks!
Thanks for your question. Perhaps a different metaphor will help. Let's say you want to cook a meal and you want Chat GPT to give you the recipes. You tell Chat that you like Italian, French, Mexican and Thai food. The problem: is there lots of types of Italian food, French food... etc. So, you then give Chat examples of your favorite recipes for each of these food types. Now, ChatGPT REALLY knows what you like when you say that like Italian food (it's no longer a broad and generic statement). Essentially, there's a big difference between "I like food" and "I like this meal, with these ingredients, cooked this way." Once you've primed Chat GPT with that information, your outputs will be much more specific to your needs.
The main issue with Chat is that it can't digest (see the pun I used there) all of the recipes ...from all of your favorite food types in one shot. So, you use variables to enable it to store ALL of your information before seeking a response. I hope that helps to clarify.
Why couldn't I just say store these different examples of comedy and randomly draw from their styles to write some comedy using the following components?
You could certainly do that. However, chat only has so much memory per input. So, if you enter 5000 words, the comedy example at the top of the input may get full consideration while the example at the bottom of the input may not be reviewed by ChatGPT at all. If you input the data in chunks however (and label those chunks), ChatGPT has the opportunity to consume all of the examples you'd like it to reference. Essentially, this approach is just a hack to maneuver within its limited system memory until Chat can manage more tokens per dialogue. Make sense?