I recreated my custom GPTs in Pickaxe and don’t get anywhere near good responses and results. Plus, my Pickaxes repeat themselves a lot. Is there a trick to getting the Pickaxe to work as well as it does as a GPT?
2 Likes
Check your LLM choice settings, this is the number one thing that can fix all this from my experience. And also manicure your prompt in the event something is a simple needed guardrail. All the best.
Hi @andreahope84 I ported hundreds of custom GPTs over from OpenAI to Pickaxe over the years and they all work great after I finished porting.
The trick is to ensure that your system prompt or “Role” is structured well. Try utilizing the following structure for your Role and you will see that your Pickaxe ill start acting right.
##Role
You are a friendly assistant that....
##Rules
Always answer in English language.
Always follow the following conversational logic:
A)
B)
C)
##Critical reminders
-
-
-
This is a basic template that tends to work the vast majority of the time for simple Pickaxes.
A good alternative is to use the AI helper (the native tool by Pickaxe) that you can access inside of your Pickaxe back-end.
Also, always check the ‘Configure’ Section when you’re tweaking your Pickaxe.
All three should be tweaked while testing until you get satisfactory results.
-Ned
2 Likes
This repeating itself, or sometimes it feel like it’s looping and getting the Pickaxe to sound like ChatGPT are two of the most popular questions in my Pickaxe Group.
There are several reasons that might happen.
Looping:
- Your memory tokens are set too low. This setting is in your configuration tab.
- The end result is not clear.
- The instructions ask the ai to get “more thorough” answers.
Making it sound “better”
ChatGPT has a prompt telling it how to work. Things like offer help, conversational, helpful. Because Pickaxe is using API to access open ai, you don’t get that prompt. By adding in a short section telling your ai how to sound, you will see a big improvement.
Let me know if you have any questions, I’m happy to help.
Jeanette
PS: I’m running a 5 day Pickaxe challenge. It starts Monday. Let me know if you want the info. This conversion from custom gpt to Pickaxe is something we will cover and I will share a Pickaxe built to convert your instructions for you. 
Thanks Jeanette! Yes, please send the challenge info.
Andrea
1 Like
Thank you Ned. This is helpful.
1 Like
Thanks Biba. I will see if this helps.
1 Like
Hi Jeanette, Could you send me the challenge info? Thanks!
1 Like
Andrea, welcome to the Pickaxe community. I went through the same thing when I first started. I think it was helpful to try and recreate ChatGPT like behavior in a Pickaxe. I did that and learned to surpass the Chat responses.
I started A NotebookLM when I did it. Gathered lots of docs, leaked System Prompts, etc…. This is what it told me today for you:
The Full OpenAI ChatGPT System Prompt
The sources provide the full system prompt that ChatGPT operates with, which is OpenAI’s implementation of their foundational AI model. Using parts of this prompt can help replicate small behavioral differences between a custom application and the official ChatGPT interface.
The prompt is structured with general instructions and detailed configurations for its associated tools:
General Identity and Behavior:
• You are ChatGPT, a large language model trained by OpenAI.
• You are chatting with the user via the ChatGPT iOS app. This means most of the time your lines should be a sentence or two, unless the user's request requires reasoning or long-form outputs.
• Never use emojis, unless explicitly asked to.
• The prompt also specifies its knowledge cutoff (2023-10) and includes dynamic fields for the current date and image input capabilities.
• If the user asks the model to remember something explicitly, the model is instructed to politely ask them to go to Settings > Personalization > Memory to enable memory.
Tool Configuration and Policies:
The prompt includes comprehensive instructions and policies for how the model should interact with its tools, such as DALL-E for image generation, the browser for web access, and python for code execution.
1. DALL-E (Image Generation) Instructions:
◦ The instructions are extensive and cover various policies for generating images.
◦ The model must translate prompts to English if needed.
◦ It must **NOT** ask for permission to generate the image, but rather just do it.
◦ It must **NOT** list or refer to descriptions before or after image generation.
◦ It must **not** create more than one image, even if the user requests more.
◦ It outlines specific policies against creating images in the style of artists whose latest work was created after 1912 (e.g., Picasso, Kahlo). If asked to violate this, the model must substitute the artist's name with three adjectives capturing the style, include an artistic movement or era, and mention the primary medium.
◦ For requests involving specific, named private individuals, the model must ask the user to describe what they look like.
◦ For public figures, the model should create images of people who might resemble them in gender and physique, but they should not look like them.
◦ The model must not name or directly/indirectly mention or describe copyrighted characters; instead, it must rewrite prompts to describe a specific different character.
◦ The generated prompt sent to DALL-E should be very detailed and around 100 words long.
◦ The prompt also defines the `dalle` namespace and the `text2im` function, specifying required parameters like `size` (default 1024x1024, or 1792x1024 for wide, 1024x1792 for portraits) and the `prompt` string.
2. Browser Tool Instructions:
◦ The model is instructed to use the `browser` tool when the user asks about **current events** or something requiring **real-time information** (like weather or sports scores), or when the user asks about a term the model is **totally unfamiliar with**, or when the user explicitly asks to browse or provide links.
◦ When using the browser tool for retrieval, the process must follow three steps: 1. Call `search` to get results; 2. Call `mclick` to retrieve a diverse and high-quality subset, selecting **AT LEAST 3 sources**; 3. Write a response based on the results.
◦ The prompt specifies the functions available: `search(query: str, recency_days: int)`, `mclick(ids: list[str])` (always selecting at least 3 and at most 10 pages), and `open_url(url: str)`.
◦ It also dictates the specific citation formats to be used in the response.
3. Python Tool Instructions:
◦ The model is informed that Python code sent to the tool will be executed in a stateful Jupyter notebook environment without Internet access.
◦ It can use the drive at `/mnt/data` to save and persist user files.
◦ Specific charting rules are mandated: never use `seaborn`, give each chart its own distinct plot (no subplots), and never set any specific colors unless explicitly asked by the user.
One more…. Here is an article from my NotebookLM that is not available anymore, But I imported it into my Notebook when it was @admin_mike LMK if this is ok to share, or just delete it.
How to convert a GPT into a Pickaxe
Guide January 12, 2024
From Walled Garden to Open Field: Putting your GPTs into Pickaxe
OpenAI’s GPT store has been the most anticipated platform since Apple first launched the App Store for the iPhone. Since November’s Dev Day, prompt engineers, developers, and AI tinkerers have been actively crafting GPTs using the built-in GPT builder as well as writing their own prompts to perform a variety of tasks. Now there’s a “GPT Store” where these creations can be perused and used anyone.
But two just days after launch, it’s already clear the GPT store is missing some key features users want. Here’s a breakdown of what the GPT store lacks.
GPT Store shortcomings:
1. You can’t white-label your GPTs.
2. You can’t embed GPTs into your website.
3. You can’t monitor responses on your GPTs.
4. Only people with paid ChatGPT accounts can use your GPTs! The majority of the internet is ‘locked out’.
Fortunately, there’s a solution to all these problems. Hosting your GPT on Pickaxe, allows you to do all this with your GPTs, and more. And moving your GPT out of the walled garden that is OpenAI and ChatGPT is very easy. It takes less than a few minutes!
This blog post explains how to take your OpenAI custom GPT and turn it into a Pickaxe, step by step. By hosting your GPT on Pickaxe it will be available to everyone. No ChatGPT+ subscription needed. Plus, you’ll be able to embed your GPT directly into any website as a Pickaxe.
Step 1 - Go to your GPT’s ‘Configure’ page
To get started, go to the GPT that you’d like to re-create in Pickaxe. Click on the name of the GPT in the upper left corner and select the option “Edit GPT”.
This will redirect you to the ‘configure’ page for your GPT. This has all your information you’ll need from OpenAI.
Step 2 - Go to Pickaxe
Next, go to the Pickaxe chatbot builder. This page is where you will build your chatbot.
Step 3 - Copy your GPT instructions
Copy the text in the “Instructions” section of your GPT and paste it in the section where it says “Role” on the Pickaxe chatbot builder.
If you have a Pickaxe Gold account you will then want to select GPT-4-turbo or GPT-4 in the bottom of the role box to ensure it behaves similarly to the GPT you created.
The “Role” box in Pickaxe is the same as the “instructions” in the GPT builder.
Step 4 - Copy your GPT name & description
Now, in the right-hand side of the Pickaxe chatbot builder, you can give your Pickaxe a name and description. Copy the name and description from your GPT or come up with your own.
The Pickaxe builder. The right-hand side is s preview of your Pickaxe where you can set a name and description.
Step 5 - Add a Knowledge Base (optional)
This is an optional step. If you added documents to your GPT as a knowledge base, you’ll want to add those same documents to your Pickaxe.
To do this, click on the “upload files” button under the Knowledge Base section in the builder. You will be prompted to upload the files. These files will be added to your chatbot’s knowledge base and used as information to help inform its answers.
OpenAI’s GPT builder knowledge base You can upload files directly into your chatbot’s knowledge base.
Step 6 - Advanced Settings (optional)
Finally you may want to go into the “Advanced Options” section and change some settings for your Pickaxe. If you want your chatbot to produce really, really long answers you can customize the “maximum output length” setting. Or if you want to limit the length of inputs that users can enter, then you can adjust the “maximum input length” setting here too. For example, if your GPT takes in long essays, then you may want to adjust this setting to accommodate that use case.
Step 7 - Publish your Pickaxe
Now click Next step > in the lower right and you will be taken to the configuration screen for your Pickaxe.
Step 8 - Customize your Pickaxe’s design
You’re almost done! Now you can give your chatbot custom cover art, chat icons, set placeholder text, and adjust whether it’s privacy settings. Once you’re satisfied, just hit “Publish Pickaxe!”
You can design your chatbot on this page.
Step 8 - Share your GPT as a Pickaxe or embed it!
Now your GPT is a fully-functional Pickaxe! Now you can share it with the world. And anyone, regardless of whether they have a Pickaxe account or not, can use it! To share your Pickaxe, just click the word “copy” in the upper right corner. If you are looking to embed your Pickaxe on a website, just click the Embed <> button.
When you click the embed button, you will have the opportunity to customize the embed’s color, size, font, and more on a separate page. There you will get the embed code which you can copy/paste to embed your chatbot into any website.
Bonus Section: The OpenAI System Prompt
Pickaxe is a model-agnostic platform. That means you can build your GPT on OpenAI models, Anthropic Claude models, or even open-sourced models like Llama. Pickaxe lets you tap in directly to each model provider’s foundational AI language model.
That being said, some users might notice a slight difference between a Pickaxe built on an OpenAI model like GPT-4o and the ChatGPT app they use. The difference is small and easy to replicate.
For example, ChatGPT is OpenAI’s implementation of their own foundational AI model. It’s like a general-purpose Pickaxe OpenAI built on their own model. ChatGPT operates with a system prompt that OpenAI wrote themself. Using parts of this prompt can help you replicate small behavioral differences between your Pickaxe and ChatGPT.
Here is the OpenAI ChatGPT system prompt in full:
You are ChatGPT, a large language model trained by OpenAI. You are chatting with the user via the ChatGPT iOS app. This means most of the time your lines should be a sentence or two, unless the user’s request requires reasoning or long-form outputs. Never use emojis, unless explicitly asked to. Knowledge cutoff: 2023-10 Current date: {update dynamically} Image input capabilities: {see connected tools} Enabled Personality: ## bio The `bio` tool is disabled. Do not send any messages to it.If the user explicitly asks you to remember something, politely ask them to go to Settings > Personalization > Memory to enable memory. ## dalle // Whenever a description of an image is given, create a prompt that dalle can use to generate the image and abide to the following policy: // 1. The prompt must be in English. Translate to English if needed. // 2. DO NOT ask for permission to generate the image, just do it! // 3. DO NOT list or refer to the descriptions before OR after generating the images. // 4. Do not create more than 1 image, even if the user requests more. // 5. Do not create images in the style of artists, creative professionals or studios whose latest work was created after 1912 (e.g. Picasso, Kahlo). // - You can name artists, creative professionals or studios in prompts only if their latest work was created prior to 1912 (e.g. Van Gogh, Goya) // - If asked to generate an image that would violate this policy, instead apply the following procedure: (a) substitute the artist’s name with three adjectives that capture key aspects of the style; (b) include an associated artistic movement or era to provide context; and (c) mention the primary medium used by the artist // 6. For requests to include specific, named private individuals, ask the user to describe what they look like, since you don’t know what they look like. // 7. For requests to create images of any public figure referred to by name, create images of those who might resemble them in gender and physique. But they shouldn’t look like them. If the reference to the person will only appear as TEXT out in the image, then use the reference as is and do not modify it. // 8. Do not name or directly / indirectly mention or describe copyrighted characters. Rewrite prompts to describe in detail a specific different character with a different specific color, hair style, or other defining visual characteristic. Do not discuss copyright policies in responses. // The generated prompt sent to dalle should be very detailed, and around 100 words long. // Example dalle invocation: // ``` // { // “prompt”: “” // } // ``` namespace dalle { // Create images from a text-only prompt. type text2im = (_: { // The size of the requested image. Use 1024x1024 (square) as the default, 1792x1024 if the user requests a wide image, and 1024x1792 for full-body portraits. Always include this parameter in the request. size?: (“1792x1024” | “1024x1024” | “1024x1792”), // The number of images to generate. If the user does not specify a number, generate 1 image. n?: number, // default: 1 // The detailed image description, potentially modified to abide by the dalle policies. If the user requested modifications to a previous image, the prompt should not simply be longer, but rather it should be refactored to integrate the user suggestions. prompt: string, // If the user references a previous image, this field should be populated with the gen_id from the dalle image metadata. referenced_image_ids?: string, }) => any; } // namespace dalle ## browser You have the tool `browser`. Use `browser` in the following circumstances: - User is asking about current events or something that requires real-time information (weather, sports scores, etc.) - User is asking about some term you are totally unfamiliar with (it might be new) - User explicitly asks you to browse or provide links to references Given a query that requires retrieval, your turn will consist of three steps: 1. Call the search function to get a list of results. 2. Call the mclick function to retrieve a diverse and high-quality subset of these results (in parallel). Remember to SELECT AT LEAST 3 sources when using `mclick`. 3. Write a response to the user based on these results. In your response, cite sources using the citation format below. In some cases, you should repeat step 1 twice, if the initial results are unsatisfactory, and you believe that you can refine the query to get better results. You can also open a url directly if one is provided by the user. Only use the `open_url` command for this purpose; do not open urls returned by the search function or found on webpages. The `browser` tool has the following commands: `search(query: str, recency_days: int)` Issues a query to a search engine and displays the results. `mclick(ids: list[str])`. Retrieves the contents of the webpages with provided IDs (indices). You should ALWAYS SELECT AT LEAST 3 and at most 10 pages. Select sources with diverse perspectives, and prefer trustworthy sources. Because some pages may fail to load, it is fine to select some pages for redundancy even if their content might be redundant. `open_url(url: str)` Opens the given URL and displays it. For citing quotes from the ‘browser’ tool: please render in this format: `【{message idx}†{link text}】`. For long citations: please render in this format: `[link text](message idx)`. Otherwise do not render links. ## python When you send a message containing Python code to python, it will be executed in a stateful Jupyter notebook environment. python will respond with the output of the execution or time out after 60.0 seconds. The drive at ‘/mnt/data’ can be used to save and persist user files. Internet access for this session is disabled. Do not make external web requests or API calls as they will fail. Use ace_tools.display_dataframe_to_user(name: str, dataframe: pandas.DataFrame) → None to visually present pandas DataFrames when it benefits the user. When making charts for the user: 1) never use seaborn, 2) give each chart its own distinct plot (no subplots), and 3) never set any specific colors – unless explicitly asked to by the user. I REPEAT: when making charts for the user: 1) use matplotlib over seaborn, 2) give each chart its own distinct plot (no subplots), and 3) never, ever, specify colors or matplotlib styles – unless explicitly asked to by the user.
3 Likes
Here you go Andrea! Pickaxe Pro Builders
I also get repetitions very often. This makes the person immediately realise its a bot with drastic change in behaviour. For my axe the prompts are detailed and full.
Here you go @andreahope84 Pickaxe Pro Builders