Hi everyone,
We recently launched the OpenAI command pack. It allows you to auto-generate any text anywhere.
If you'd like to use it without signing up for OpenAI, please reach out to me (dan@blaze.today)
Please use this thread to share snippets that use the command pack.
The snippet with the most likes wins bragging rights.
Copy these snippets to your dashboard to make them work
This snippet translates text to any language:
{note} Text to translate:
{formparagraph: name=text; cols=40; rows=10}
Translate to {formmenu: default=Spanish; Portuguese; Mandarin Chinese; French; German; Arabic; Hindi; English; Japanese; Bengali; Russian; name=language}
{endnote}
{openai-text: Translate the following to {=language}:
{=text}; trim=left}
This one simply rewrites your text to make it more clear and fixes typos and grammatical errors:
{note} Text to rewrite:
{formparagraph: name=text; cols=40; rows=10}{endnote}
{formtoggle: name=Generate response; default=no; trim=left}
{openai-text: Rewrite the following to make it more clear and fix typos and grammatical errors:
{=text}; trim=left}{endformtoggle}{if: `generate response`=="no"}{error: Check "Generate response" to create the rewritten text; block=yes}{endif}
Generate a summary of the content of your clipboard. Copy any text, article, transcript, etc. and use this snippet to create a summary
{openai-text: Summarize the following text in bullet points:
{clipboard}; trim=left}
Writing a personal connection request on LinkedIn:
{openai-text: write a connection request to {linkedin-firstname} on LinkedIn (max 300 characters). Tell them why I love their company. Here is some information about {linkedin-firstname}: {linkedin-headline}}
Respond to a Gmail email
This snippet responds to the last email in a Gmail thread. You can tell it what content to include in the response and what tone to use:
{note}{`original email`=replace({site: text; multiple=yes; selector=[data-message-id*=\"#msg-\"] .ii:not([style=\"display:none\"]) .a3s.aiL}[1], "\n\n", "\n")}
Items to include in the response:
{formparagraph: name=raw response; default=; cols=40; rows=8}
{if: `raw response`==""}{error: Consider adding content to include in the response; block=no}{endif}
Tone to use: {formmenu: professional; default=casual; kind; funny; polite; name=tone; multiple=yes; cols=40}
Final prompt:
{prompt="Write an email to " & {gmail-firstname} &" in a "& {=join(tone, "BLAZE_AND")}&" tone, responding to the following email:\"" &{=`original email`} &"\".\n Include the following in the respose: " & {=`raw response`} & "."}{formparagraph: name=prompt; cols=40; rows=10}{endnote: trim=yes}{formtoggle: name=generate response; default=no}{openai-text: {=prompt}}{endformtoggle}{if: `generate response`=="no"}{error: Check "Generate response" to create the response; block=yes}{endif}
Generic prompt
Use this to type any prompt manually.
{note} {formparagraph: name=prompt; cols=40; rows=10}{endnote}
{formtoggle: name=Generate text; default=no; trim=left}
{openai-text: {=prompt}; trim=left}{endformtoggle}{if: `generate text`=="no"}{error: Check "Generate text" to create the text; block=yes}{endif}
Gmail Thread Summarizer
If every morning you have to read long email threads to catch up on what happened, then this snippet is for you:
{note}
Toggle configuration/generation: {formmenu: default=Configure; Generate; name=current_state}{endnote}{note: preview=yes; insert=no}
{endnote}{if: current_state=="Configure"}{note}
{contexts={site: text; multiple=yes; selector=[data-message-id*="#msg-"] .ii:not([style="display:none"]) .a3s.aiL}}
Choose previous contexts: {formmenu: multiple=yes; values={=contexts}; name=chosen_contexts}
{previous_emails_str=join(map(chosen_contexts, (x, i) -> "\nEmail " & i & ": \"" & x & "\""), "\n"); trim=yes}
{`mail string`=previous_emails_str; trim=yes}
{`mode string`="Please only write a short summary for the email thread given in following lines. Do not complete the email thread."; trim=left}
Final prompt:
{prompt=`mode string` & "\n" & `mail string`}{formparagraph: name=prompt; cols=70; rows=15}
{error: Use the dropdown above to "Generate" text; block=yes}
{endnote: trim=yes}{else}{openai-text: {=prompt}; trim=yes}{endif}
This is how it looks when used:
I just used it in a Gmai thread with 11 back and forth emails. The summary I received is:
Summary: [user] asked Andrew, Head of Customer Success at Text Blaze, if it was possible to add an existing folder in a drop down form menu field. After discussing the issue over Zoom, Andrew provided a solution and [user] implemented it successfully. Andrew then provided further assistance when [user] attempted to apply the same solution to a different issue.
...which, of course, is a very accurate summary. You can tune the prompt to request for "short summary", "detailed summary", etc.
You can also modify the snippet to summarize other conversations, like Slack message threads, Outlook email threads, Salesforce comments, etc. If you need help doing that, feel free to message me.
Dan this is impressive and extremely useful.
Keystroke shortcut would really add functionality
George
I used one for the fun of it the other day and wound up making a snippet of the text that was returned 
{openai-text: write an empathy statement for not having solved an issue in a timely manner.; format=text; name=empathy}
This returned: I understand how frustrating it must be to have waited for a resolution to this issue for so long. I apologize for not being able to provide a solution in a more timely manner.
This stuff is both existing and scary at the same time. Awesome that you've integrated it so well.
This works really well!
I also was able to have it generate a summary of a webpage by adding the link into the snippet, but I wasn't able to get it by putting the link in formtext field. Do you know a good way to do that? I suppose that this snippet might actually work if you copy the link, but I haven't tried that yet.
The currently available Open AI API appears to be GPT-3 and not Chat GPT. The amount of training and the model is different, so Chat GPT is smarter, so the responses coming from the current API are a bit stupid.
They are actually from the same 3.5 model. The difference is you must know a bit more about GP3 to use it directly through an interface or the API. ChatGPT does this under the covers by setting the parameters based on your conversational style and other factors in the prompts. You can get high-quality results from both.
Unknown to the user, ChatGPT is also a bit of an expert prompt engineer. What you type is not always what the Chat interface poses to GP3.5 as the final question. Bear in mind, the chat system is built by an AI company determined to make it perform in the best light and they are using AI to enhance your experience.
We are free to do the same. Often, my Text Blaze recipes send a number of prompts to GP3 before showing the user the final result. Some use 7 queries to get to reliable and useable answers.
Lastly, if you are using GP3 for bona fide business objectives, you really should be doing so with a fine-tuned model; one that extends davinci by understanding your business domain. I have built Text Blaze-driven solutions that train a fine-tuned model every time a query is performed. The system gets smarter about things relevant to the business as the users engages with it. GP3 (out of the box is fine for experimental and learning use cases), but that's not how it should be used for everyday intelligence support.
@Bill_French
Thank you, I am an internist, so I guess I don't have a proper understanding of the technical background.
GPT4 has been released and the integration of AI into Text Blaze could be a revolutionary leap forward.
Interesting. I thought the latest access was a 3.5 variant.
I know some of you have already tried it, but Bing Chat（GPT4) is awfully smart. It replies in Japanese and if Chat GPT is 10 years old, it's like 20 years old. And it's only been a month since Chat GPT was released. Singularity seems to be coming.