Dgreenwood: Created page with "{{AutoVersion}}
{{#lst:Glossary|Ask AI}}
{|class="download-box" | File:Asset 22@4x.png | You may download the ZIP(s) below and upload it into your own Grooper environment (version 2024). The first contains one or more '''Batches''' of sample documents. The second contains one or more '''Projects''' with resources used in examples throughout this article. * Media: 2024_Wiki_Ask-AI_Batch.zip * Media: 2024_Wiki_Ask-AI_Project.zip |}..."

2025-08-27T21:09:53Z

Created page with "{{AutoVersion}} <blockquote>{{#lst:Glossary|Ask AI}}</blockquote> {|class="download-box" | File:Asset 22@4x.png | You may download the ZIP(s) below and upload it into your own Grooper environment (version 2024). The first contains one or more '''Batches''' of sample documents. The second contains one or more '''Projects''' with resources used in examples throughout this article. * Media: 2024_Wiki_Ask-AI_Batch.zip * Media: 2024_Wiki_Ask-AI_Project.zip |}..."

New page

{{AutoVersion}}

<blockquote>{{#lst:Glossary|Ask AI}}</blockquote>

{|class="download-box"
|
[[File:Asset 22@4x.png]]
|
You may download the ZIP(s) below and upload it into your own Grooper environment (version 2024). The first contains one or more '''Batches''' of sample documents. The second contains one or more '''Projects''' with resources used in examples throughout this article.
* [[Media: 2024_Wiki_Ask-AI_Batch.zip]]
* [[Media: 2024_Wiki_Ask-AI_Project.zip]]
|}

== About ==
'''''Ask AI''''' is a '''''Value Extractor''''' introduced in '''[[https://wiki.grooper.com/index.php?title=What%27s_New_in_Grooper_2024|Grooper 2024]]'''. It is a replacement for the now outdated '''''GPT Complete''''' extractor.
* '''''GPT Complete''''' uses a deprecated method to call the OpenAI [https://en.wikipedia.org/wiki/API API].
* '''''Ask AI''''' can be used with any [https://en.wikipedia.org/wiki/Large_language_model LLM] Service Provider.
* '''''Ask AI''''' has other advantages over '''''GPT Complete'''''
The general idea behind '''''Ask AI''''' is simple: prompt an LLM [https://en.wikipedia.org/wiki/Chatbot chatbot] with a question about a document to return a result. With appropriate instructions, '''''Ask AI''''' can even parse [https://en.wikipedia.org/wiki/JSON JSON] responses into a '''Data Model's''' instance hierarchy. For example, you can use '''''Ask AI''''' as a '''''Row Match''''' '''''Table Extract Method'''''.
<div style="padding-left: 1.5em;">
=== Ask AI Pros and Cons ===
<big>Pros</big>
* Returns data using a natural language prompt.
* Less knowledge of Grooper extractors required to return data
* Quicker time to value.
* Easier to maintain over time.

<big>Cons</big>
* LLM responses can be unpredictable.
* LLM responses can be inaccurate.
* LLMs answer by predicting the next best word one after another. They do not “know” anything.
* As an extractor, “Ask AI” must be configured “per field”.
* An API call is made every time the extractor executes.
** This is '''''<span style="color: red;">very</span>''''' important to consider. Limiting the text sent to the LLM whenever possible should be a consideration. The '''''Context Extractor''''' property can play a big role in helping with this.
* No result highlighting

=== Properties ===
[[image: 2024_Ask-AI_00_01.png]]

<big>''Model''</big><br>
The API Key you use will determine which GPT models are available to you. The different GPT models can affect the text generated based on their size, training data, capabilities, prompt engineering, and fine-tuning potential. [https://openai.com/index/gpt-3-apps/ GPT-3's] larger size and training data, in particular, can potentially result in more sophisticated, diverse, and contextually appropriate text compared to [https://openai.com/index/gpt-2-1-5b-release/ GPT-2]. However, the actual performance and quality of the generated text also depend on various other factors, such as prompt engineering, input provided, and specific use case requirements. GPT-4o is the latest version, as of this writing, and takes the GPT model even further.

<big>''Parameters''</big><br>
Please see the [[Parameters (Property)|Parameters]] article for more information.

<big>''Instructions''</big><br>
The instructions or question to include in the prompt. The prompt sent to OpenAI consists of text content from the document, which provides context, plus the text entered here. This property should ask a question about the content or provide instructions for generating output. For example, "what is the effective date?", "summarize this document", or "Your task is to generate a comma-separated list of assignors".

<big>''Preprocessing''</big><br>
Please visit the [[Text Preprocessor]] article for more information.

<big>''Context Extractor''</big><br>
An optional extractor which filters the document content included in the prompt. All '''''Value Extractor''''' types are available.

<big>''Max Response Length''</big><br>
The maximum length of the output, in tokens. 1 token is equivalent to approximately 4 characters for English text. Increasing this value decreases the maximum size of the context.

<big>''Parse JSON Response''</big><br>
If this property is enabled, JSON returned in the response will be parsed into a Data Instance hierarchy.

Use this mechanism to capture complex data and generate output instances with named children, producing output similar to using named regex groups with '''''Pattern Match''''', or using '''''Ordered Array''''' collation. This type of Data Instance hierarchy can be consumed by the '''''Row Match''''' '''''Table Extract Method''''' or the '''''Simple''''' '''''Section Extract Method'''''.

When this property is enabled, this instructions should ask the AI to respond with JSON, and provide instructions and examples as need to ensure the AI understands the desired JSON format. The JSON may contain a single JSON object or an array of JSON objects.

If a single JSON object is returned, a single output instance will be generated, containing one named child for each property of the JSON object. This type of output would be appropriate for capturing a single-instance Data Section.

If a JSON array is returned, one output instance will be generated for each object in the array. Each output instance will have named children reflecting the properties of the JSON object. This type of output is appropriate for capturing a '''Data Table''' or multi-instance '''Data Section'''.

The AI must respond with JSON only, or with the JSON delimited using the prefix and suffix shown below. OpenAI models are typically trained for this out of the box, but models trained by other organizations may require special instructions.
</div>
== How To ==
<div style="padding-left: 1.5em;">
=== Establish an LLM Connector ===
First, we need to establish an '''''LLM Connector''''' within the '''''Options''''' property on the '''Root''' object.

Please visit the '''''[[LLM Connector (Repository Option)|LLM Connector]]''''' article for more information.

=== Configure Ask AI ===
<div style="padding-left: 1.5em;">
==== Buyer ====
# Select the "buyer" '''Data Field'''.
# Click the drop-down button for the '''''Value Extractor''''' property.
# Select ''Ask AI'' from the drop-down menu.

[[image: 2024_Ask-AI_01_01.png]]

# <li value=4>With the '''''Value Extractor''''' property set, click the ellipsis button to configure the extractor.</li>

[[image: 2024_Ask-AI_01_02.png]]

# <li value=5>In the "Value Extractor" window click the ellipsis button for the '''''Model''''' property.</li>
# In the "Model" window select ''gpt-4o''. As of the writing of this article, it is the most accurate model, and the best choice.

[[image: 2024_Ask-AI_01_03.png]]

# <li value=7>Click the ellipsus button on the '''''Instructions''''' property.</li>
# Add a prompt in the "Instructions" window.<pre>Who is the buyer? Respond with just the name.</pre>

[[image: 2024_Ask-AI_01_04.png]]

# <li value=9>In the '''Batch Viewer''' be sure to select the document from the provided '''Batch'''.</li>
# If the toggle for auto extraction is on, you will automatically get a result. It may be a good practice to turn this off when testing to limit token consumption.
#* Speaking of limiting token consumption, this may be a good opportunity to look at the '''''Context Extractor''''' property. Knowing that the "buyer" name is almost certainly always listed on the first page, you could make an extractor that only returns the text from the first page. Since this document is 3 pages, limiting the text sent to the AI would greatly reduce token consumption.
# Observe the returned value, but take note there is no highlighting.

[[image: 2024_Ask-AI_01_05.png]]

==== Seller ====
# Select the "seller" '''Data Field'''.
# Set the '''''Value Extractor''''' to ''Ask AI'' and click the ellipsis button.

[[image: 2024_Ask-AI_02_01.png]]

# <li value=3>Set the '''''Model''''' property to your choice of model.</li>
# Add a prompt in the '''''Instructions''''' property.<pre>Who is the seller? Respond with just the name.</pre>
# In the '''Batch Viewer''' select the document from the provided '''Batch'''.
# Use auto extract or manually run extraction.
# Observe the result in the '''Results''' list view.

[[image: 2024_Ask-AI_02_02.png]]

==== Summary ====
# Select the "summary" '''Data Field'''.
# Set the '''''Value Extractor''''' to ''Ask AI'' and click the ellipsis button.

[[image: 2024_Ask-AI_03_01.png]]

# <li value=3>Set the '''''Model''''' property to your choice of model.</li>
# Add a prompt in the '''''Instructions''''' property.<pre>Give a summary in 50 words or fewer.</pre>
# Select the document from the provided '''Batch'''.
# Use auto extract or manually run extraction.
# Observe the result in the '''Results''' list view.

[[image: 2024_Ask-AI_03_02.png]]

==== Lands and Leases ====
# Select the "LandsAndLeases" '''Data Table'''.
# Set the '''''Extract Method''''' property to ''Row Match''.
# Click the drop-down for the '''''Row Extractor''''' property and select ''Ask AI''.

[[image: 2024_Ask-AI_04_01.png]]

# <li value=4>Click the ellipsis button on the '''''Row Extractor''''' property.</li>

[[image: 2024_Ask-AI_04_02.png]]

# <li value=5>Set the '''''Model''''' property to your choice of model.</li>
# Add a prompt in the '''''Instructions''''' property.
#* <pre>What is the content of the "Additional Lands and Leases" table? Answer with JSON formatted as below. The quote property should contain a quote from the document which includes each individual full row: [{section: "", township: "", range: "", county: "", state: "", quote: ""}]</pre>
#* Notice how this prompt has been engineered to return values as JSON objects. These objects will be created as sub-elements of the main extracted result. Because the provided JSON in the prompt defines the objects using the exact names of the '''Data Columns''', the returned sub-elements will be able to populate their respective '''Data Columns'''.
# Enable the '''''Parse JSON Response''''' property.
#* This property must be enabled to take advantage of the parsed JSON objects as sub-elements described in the previous step.
# Select the document from the provided '''Batch'''.
# Use auto extract or manually run extraction.
# Observe the result in the '''Results''' list view.

[[image: 2024_Ask-AI_04_03.png]]

# <li value=11>Select the '''Data Model''' from the provided '''Project'''.</li>
# Click the "Tester" tab.
# Select the document from the provided '''Batch''' in the '''Batch Viewer'''.
# Click the "Test" button.
# View the extracted results in the '''Data Model Preview'''.

[[image: 2024_Ask-AI_04_04.png]]
</div>
</div>

2024:Ask AI (Value Extractor) - Revision history