667554_en-US

liliconcom

Re: AI Assistant

Update as of 8/16: 

 

One observed issue is that when data tables with scripts attached are sent to models, the attached scripts take up a lot of tokens. So, the conversation use too many tokens and throw an error pretty quickly. Ways to get around this are:

1. raise the token cap (based on what the model allows)

2. copy the data table and delete the scripts from that version, and send that to the model

3. reset the conversation or restart the add-in whenever the conversation might be too long and an error throws

 

I will add a token display in the near future to provide to clarity on how long a conversation has been while using the add-in to hopefully avoid this confusion, and in the meantime, hopefully these suggestions should solve that problem.

Re: AI Assistant

Hi Parker,

 

I like your add-in, and I am hoping to try it out.  Is there a particular version of Python we should be using?  Does your add-in work with JMP 17?

 

Kind regards,

Bill

Re: AI Assistant

Hi @Bill_Worley ,

I've been using 3.11, but it should work on older versions of Python. The add-in also works with JMP 17, although you'll have to install the OpenAI module on your computer. Let me know if something doesn't work or you have any questions!

 

Cheers,

Parker

Re: AI Assistant

we have our own ChatGPT hosted behind a firewall and I'm not sure how to connect our AI to this add-in. How do I get a Key if it's hosted like this?

AI Assistant

Note: This add-in Requires JMP 17 or later. This add-in also requires the openai module to be downloaded for the version of Python JMP is using, although the original version only requires openai. This add-in also incurs API usage expenses to the user through their personal openai account. Please read the write-up before using the add-in as some important details for use are included. As this add-in was originally intended to be a proof-of-concept, we are saying that development and maintenance for this add-In is complete.

 

Known issues:

• Please Note: generated text and code can often be incorrect or produce errors. The add-in uses OpenAI's models, not JMP-specific AI. Please use accordingly.
• Token usage and limits can be finnicky. If the add-in repeats a response or throws an argument error, restart the conversation and/or add-in.
• The tables and reports will only update once the add-in is opened if a query is submitted. To avoid issues, open up tables or reports before opening the add-in, or submit a query after opening up a new one.
• Switching from "do it for me" to "explain and assist" can sometimes produce errors. If this happens, restart the add-in.

 

Overview:

This addin is a live prototype of leveraging generative AI to enhance the JMP experience. Connected to OpenAI’s API, this addin features a conversational chatbot interface with JMP-specific features:

1. Querying about data tables or reports

2. A mode for instructing the model to execute tasks in JMP (and seeing the JSL it wrote)

3. Prompt libraries for storing useful queries for easy access

Users are required to bring their own API key to access the OpenAI models powering the add-in and might have to install the openai Python module on their machine. Please note that the models aren't always accurate. In other words, it can and inevitably will write incorrect JSL code, misunderstand instructions, produce wrong information, etc. Please share any feedback and results from using this add-in; we plan on making improvements soon!

 

Capabilities:

1. Explain & assist mode: type a query into the query box, hit go, and your query, as well as the model response will appear in the conversation box.

2. Do it for me mode: type an instruction into the query box, hit go, and JMP will execute a JSL script created by the model. (If “Show Generated JSL Script” is checked, a new script window will be populated with the script written by the model)

3. Use Table/Report: send the script for an open data table or report to the model to ask subsequent queries about. (If “Hide Data in Data Tables” is checked, only the table meta data will be sent.) Please note: scripts included in tables will be sent as well, and will substantively crank up token usage if unnoticed.

4. Custom Library: prompt library with prepopulated prompts; select a category, select a prompt and hit “Use Prompt” to populate the query box. Select “Load Library” to replace the default prompts with prompts in a data table you select. (Ensure the column names are the category names and the prompts are the data)

5. Settings: enter your secret key, change the model type, token cap, and model temperature. (Information on these settings can be found here: OpenAI Platform)

6. Miscellaneous: statistics library is tasks segmented by the “Data Analysis Director” addin, the reset button will wipe the conversation history and token usage and clean the conversation box, and regenerate will submit the previously inputted query again for another response.

 

Feedback & The Future:

We're releasing this add-in in the hopes of gathering community feedback and suggestions on where to take this in the future. We think this technology has great potential, but we haven't been able to thoroughly experiment ourselves. Please send us any ideas on where to take this in the future, results from using the add-in yourself (what has worked, what hasn't worked, etc.), and your overall thoughts! We will be closely monitoring this page, and we plan on updating the add-in soon. We think this is a very exciting time for JMP, and we're looking forward to seeing your feedback!

 

Resources:

1. Video demo: 

2. OpenAI setup: How to Get an OpenAI API Key

3. OpenAI API reference: OpenAI Platform

3rd Party IntegrationAnalytical Application DevelopmentExploratory Data AnalysisRe: AI Assistant

@Bill_Worley Were you able to get to this run? It seems that we haven't provided instructions how to run/ install the openai Python module, although clues and guidance may be found in our support documentation or elsewhere on the Community.  I am getting the following error message when trying to run it: 

 

  

 

 

@d_barnett Unfortunately I am not sure in your case, and I apologize for our lack of response (as we mentioned at the top, we don't offer active Support for this Add-In, or others, although we do look to bring feature enhancements to future JMP releases, where we do support features in those releases). I attempt to loop back in @ParkerKasiewicz here to see if he can offer up any suggestions.   Thank you!     @PatrickGiuliano (JMP Technical Support)

 

Re: AI Assistant

Hi @PatrickGiuliano,

I was not able to resolve this issue. However, I believe Ryan Cooper has been able to connect to the AI Assistant.

Re: AI Assistant

Hello @PatrickGiuliano @Bill_Worley @ParkerKasiewicz ,

I get the exact same error message as @PatrickGiuliano is there any solution ?

note: I'm using v18.0.0, re-start from scratch, open Tablet Production.jmp, delete all scripts, and follow the exact same step you demonstrate on the video.

Thanks for the help.

Re: AI Assistant

@Emmanuel_Romeu None that I am aware of (I did speak with Ryan Cooper about it, but he hasn't had time to attempt to troubleshoot). I'll investigate a bit further in the coming weeks and let you know if there's any additional information I can suggest.  

Re: AI Assistant

Great, thank you @PatrickGiuliano