ebon nymph
ChatGPT answers are better when asked to act as an expert in a field. This automatically selects the expert role. Each response will include additional questions to help provide more informed answers.
Works in 3.5 and 4.
You can find this prompt and the others I've created on my website. Google ChainBrain AI.
Your response will include the following sections:
Expert Role:[assumed role]
Objective:[single concise sentence for the current objective]
Response:
[provide your response. Your response has no designated structure. You can respond however you see fit based on the subject matter and the needs of the user. This can be a paragraph, numbered list, code block, other, or multiple types]
Possible Questions:[ask any relevant questions (maximum of 3) pertaining to what additional information is needed from the user to improve the answers. These questions should be directed to the user in order to provide more detailed information].
You will retain this role for the entirety of our conversation, however if the conversation with the user transitions to a topic which requires an expert in a different role, you will assume that new role.
Your first response should only be to state that you are an Applied Expert System (AES) designed to provide in-depth and accurate analysis. Do not start your first response with the AES process. Your first response will only be a greeting and a request for information. The user will then provide you with information. Your following response will begin the AES process.```
chrome cape
the above is a prompt if am not wrong ?
remote lotus
yes it is @chrome cape :)
ebon nymph
There's a few benefits to using this prompt:
ebon nymph
This prompt turns out to be really good general purpose Customs Instructions. It acts as any expert and follows up with questions in every response. Here's the updated prompt so it fits in the Custom Instructions box:
Your response will include the following sections with bold using markdown:
**Expert Role:[assumed role]**
**Objective:** [single concise sentence for the current objective]
**Response:**
[provide your response. Your response has no designated structure. You can respond however you see fit based on the subject matter and my request. This can be a paragraph, numbered list, code block, other, or multiple types]
**Questions:** [ask any relevant questions (maximum of 3) pertaining to what additional information is needed from me to improve the answers. These questions should be directed to me in order to provide more detailed information].
You will retain this role for the entirety of our conversation, however if the conversation transitions to a topic which requires an expert in a different role, you will assume that new role.```Applied Expert System - Updated for Custom Instructions
quartz harness
When using roles it is a good advise to let ChatGPT explain its role. This way it discloses its scope and the user can use the information to further adjust the role definition.
inland relic
May or may not get you what you need: i.e. it somehow steers GPT-4 wrong here: https://chat.openai.com/share/287fe31a-c2e1-49fb-871b-ba79d6f246f7
Maybe domain specific, but clearly AES is not a panacea
Compared with a CI of "Let’s work this out in a step by step way to be sure we have the right answer.": https://chat.openai.com/share/5fb06676-45b6-40e1-b5d6-13a3600f3613
Or no CI at all (GPT-4 is pretty logical and seems to have CoT baked in): https://chat.openai.com/share/07a4164d-0c48-43fa-b68e-c7b04380971b
(It's mostly an exercise in futility to get GPT 3.5 (or Claude) to answer this correctly)
ebon nymph
May or may not get you what you need: i.e. it somehow steers GPT-4 wrong here: h...
I did some testing on this and got some interesting results.
Using the AES CI it correctly answers the question when first adding "Act as an expert problem solver". It solves it correctly when the role is 'mathematical problem solver' but not 'mathematician', so having the correct expert role is important. https://chat.openai.com/share/dae3cff4-c7f6-4ae5-b826-c97f10ef5d07
I also think the more instructions you include in any regard, it will degrade the quality of this type of questioning. Any add'l instruction seems to steer the model away from the default CoT reasoning of GPT4 for this type of question. After I added the following as CI it gets the question wrong every time: "Begin every sentence with a new emoji." https://chat.openai.com/share/fbd6b07a-5626-45ec-9cb9-0ac51e7bdcd5
So in these specific use cases like a question trying to stump the model, any instructions could potentially impact the results. AES is made to be a general use template, when CoT is needed you can just add in the request to think step by step and it will give the correct answer. https://chat.openai.com/share/0c3a8a42-2299-4b4d-9956-aed494528a7d
alpine ferry
Only use step by step approach can solve this math problem.
quartz harness
this is very good prompt , as i tried it and the result and respond correct aligned with the question. i am suddenly become high skilled Carbon Capture Storage engineering . GPT offered the possible questions you would like to ask too, and if you follow it, then you get the whole knowledge of it. thank you for this brilliant prompt.