How to use GPT-Prompter
Initial Setup
GPT-Prompter operates by using your personal OpenAI API KEY, which you can obtain by creating a profile on the OpenAI website and retrieving it from there. Rest assured that we do not have access to your key, as it is stored securely within your Chrome browser.
After installing the GPT-Prompter extension, you can find it in the list of your browser's extensions, marked by a dark icon. This icon indicates that you have not yet linked your API key to the extension. To do so, simply click on the icon, and a popup will appear, prompting you to enter your OpenAI API KEY.
To access the GPT-Prompter extension, navigate to the top right of your browser, where you will find a puzzle-shaped icon. From there, you can easily manage your extensions and utilize the GPT-Prompter's powerful capabilities, hassle-free.
Create a custom ChatGPT Prompt
In the extension popup you can create your custom Chat GPT prompt with the possibility to set:
Model: the chatGPT model "gpt-3.5 turbo" and "gpt-4"
Temperature: High means creative, low is more precise.
Max tokens: The maximum number of tokens to generate in the next message (normally one can leave it at around 1000)
Now max tokens are automatically computed from the prompt, no need to specify.
System: The assistant description, this influences the behaviour.
Finally,you can create your own user/assistant messages with the only requirement being that the placeholder #TEXT# must be included somewhere. This placeholder is replaced with the selected text from the webpage or is where the cursor will be placed if no text is selected. This feature allows for highly customizable prompts that can be tailored to your specific needs and preferences.
Create a custom Classic Prompt
In the extension popup you can create your custom prompts with the possibility to set:
Model: the completion model offered by OpenAI, from the most capable text-davinci-003 to the fastest and cheapest text-ada-001
Temperature: Higher values mean the model will take more risks. Try 0.9 for more creative applications, and 0 for ones with a well-defined answer.
Max tokens: The maximum number of tokens to generate in the completion (normally one can leave it at around 1000)
Now max tokens are automatically computed from the prompt, no need to specify.
Also here, there is the constraint that the placeholder #TEXT# must be included. This is replaced with the selected text from the webpage or is where the cursor will be placed if no text is selected.
List of custom prompts
After saving your custom prompts, they will appear in the section below along with the pre-made prompts. You can easily reorder them by dragging them to the desired location.
For each you can:
Add Title: for better sorting in the context menu. This feature can be especially useful when some prompts have long text, and it's easier to give them a title. The title will not be part of the prompt passed to GPT. Additionally, you can use the #TEXT# placeholder in the title, which will be replaced with the selected text.
Edit Prompt: this will copy the details of the prompt in the prompt design section.
Delete: you can delete any prompt that you no longer need. This feature ensures that your prompts remain organized and clutter-free, making it easy to find and use the ones you need quickly.
Furthermore, the GPT-Prompter extension offers an additional feature known as Two-Stage mode or Always open in Chat. Enabling this mode opens the prompt in the edit mode/chat, even when text is selected. This feature can be particularly useful when a prompt requires multiple steps.
In the first step, the selected text is parsed and inserted into the #TEXT# placeholder. In the second step, you can type a specific question or command in addition to the selected text, or edit the text before submission. This two-stage mode enables more complex prompts and allows for greater control over the generated output.
List of prompts
Submit the prompts to GPT API
The GPT-Prompter extension offers three primary modes: ChatGPT, Prompt On-the-Fly, and Fast Custom Prompt.
ChatGPT
An empty chat instance is opened with the system set to "You are a helpful assistant." The model "gpt-3.5-turbo" is always used in this mode.
Prompt On-the-Fly
A playground-like popup prompt that allows you to type in your prompt. You can also open the prompt after selecting text, which will immediately parse the selected text into the textarea for further editing. In this mode, you can set the temperature and any model, including "gpt-3.5-turbo".
Fast Custom Prompt
Enables you to launch a custom prompt/chats quickly. Simply highlight some text, right-click, and select the prompt you want to use from the context menu. If no text is selected, you can open the prompt and type in it before launching. Each prompt will have the first symbol corresponding to the model used, and an additional ">>" symbol to indicate if it is in Two-Stage mode. In this mode, you can also set the temperature and any model after the first launch, including "gpt-3.5-turbo".
Whisper API
It is also possible to transcribe audio from the microphone using the icon on the right side of the textbox. The user can start/stop the recording which is sent to Whisper API and then transcribed into the textbox. The language is automatically detected.
Keyboard shortcuts
The following keyboard shortcuts are available:
Alt+G: open ChatGPT floating window (this can be modified in Chrome, via the extensions page -> top left button -> keyboard shortcuts)
Alt+P: open prompt On-the-Fly (this can be modified in Chrome, via the extensions page -> top left button -> keyboard shortcuts)
Alt+Enter: Submit prompt, or regenerate/resubmit.
Esc: close prompt popup
Alt+C: Copy completion to clipboard.
Alt+A: Add completion to prompt textarea.
Advanced usage
The Advanced settings of GPT-Prompter offer additional options for further customization.
The "Show average tokens probabilities" (using logprobs) option displays the average probability of the tokens in the GPT completion along with the number of tokens. This feature can be useful in assessing the level of confidence of the model when providing the completion. However, it's worth noting that logprobs are not available for the "gpt-3.5-turbo" model by OpenAI.
The "Automatically Add completion to prompt" option works exclusively for the On-the-Fly mode. This feature automatically adds the completion text to the textarea, similar to how it's done in the OpenAI playground. This option is useful when you want to have a continuous conversation or the completion is part of a continuous text rather than a response to a specific question. However, even if this mode is not selected, you can still add the completion to the textarea manually using the "Add" button that appears after submission.
Extras
The GPT-Prompter extension provides access to a History section that logs all the prompts used along with the cost associated with each one.
In the History section, you can search through your usage history, export the prompts as JSON for fine-tuning, or delete them as needed. These features ensure that you can keep your usage organized and manageable, making it easier to optimize your workflow and get the most out of the tool.
We have recently started a Reddit group dedicated to sharing prompt examples, design tips, and ideas for using the tool to its fullest potential. Feel free to join the group and contribute your ideas and experiences!