Questions concerning AI Positron Assistant

[susannecm]

Hi everyone,
I've been playing a bit with the AI and I have some questions concerning possible use cases.
- At the moment, the assistant always seems to work on a single DITA topic. Do you have any plans to support a selection of topics, or the entire content referenced in a DITA map? Or is this not possible due to the maximum prompt size?
- Which models can be used with the assistant? All models available through the OpenAI API or only a subset? What about input and output that is not text-based?
Thank you for your help
Susanne

[Radu]

Hello Susanne,

Thanks for the feedback, please see some answers below:

At the moment, the assistant always seems to work on a single DITA topic. Do you have any plans to support a selection of topics, or the entire content referenced in a DITA map? Or is this not possible due to the maximum prompt size?

Actually our latest release of the add-on added support to refactor multiple files:
https://www.oxygenxml.com/doc/versions/ ... efactoring

- Which models can be used with the assistant? All models available through the OpenAI API or only a subset?

Not all models are available but in the Oxygen Preferences->"Plugins / Oxygen AI Positron Assistant" page you can switch to using GPT 4.
https://www.oxygenxml.com/doc/versions/ ... 5p_y14_wxb

What about input and output that is not text-based?

Not yet, we do have future plans for this though (both to allow attaching files to a request and to generate images from prompts).

[alex_jitianu]

Hello,
We released Oxygen AI Positron 2.0 in which there is a new feature to generate documentation drafts based on both text and images. Please give it a try and let us know your thoughts!
https://www.oxygenxml.com/ai_positron/whats_new.html
Our next webinar will also present this feature:
https://www.oxygenxml.com/events/2024/w ... ation.html

Best regards,
Alex

[sljohns]

I've played around with the latest release of AI positron. Well done!
Is there a way to identify the coordinates of the callouts in the image and populate an imagemap like the following for each callout? It does a good job at identifying the callouts. It just can't find the coordinates.
<area>
<shape>circle</shape>
<coords>378, 380, 14</coords>
<xref href="" format="dita"/>
</area>

[sorin_carbunaru]

Hi,

I tried to ask the AI to find the coordinates of a button from an image I attached and it just told me where in the image the button is found (e.g. bottom-left corner). But it seems it cannot compute the coordinates

[sljohns]

That's my results as well @sorin_carbunaru