Newton AI Training FAQ

Newton AI is a Virtual Artificial Intelligence powered chat assistant that can be easily added to any webpage in order to serve as an expert in your organization and automate and improve customer service. Newton AI is a human-like digital assistant that is always available to answer your user's and customer's questions in real-time, engaging with and satisfying your users like never before.

Here you can find answers to frequently asked questions as well as general information about training our customer service automation chatbot Newton AI and how our AI can be leveraged to level up your digital presence and customer service experience.


What is an Newton AI model? Newton AI is mostly powered by a type of Artificial Intelligence software called an Large Language Model, or LLM. LLMs are trained with vast sets of text data, and allow a computer program to predict what words should follow after any given user input or question. A Newton AI model is essentially a specialized module, created from text training data provided by you, that defines how your Newton AI instance will behave and respond to user's questions. When you create and train a Newton AI model, it is loaded into Chicago AI's specialized and optimized LLM platform and used to answer questions from your users.
How do I train my Newton AI instance? You can train your Newton AI instance by creating a model on the Newton AI Models portal page. When you submit the form on that page, your Newton AI LLM model will be queued for training, and will be automatically applied when training is complete.
How often do I need to train my Newton AI instance? Techically, you only have to train Newton AI once when you first provision your Newton AI instance, but you can training a new AI model any time you feel you want to add or remove data from, or tweak your Newton AI instance.
What are 'Model Prompt Rules'? 'Model Prompt Rules' allow you to provide additional instructions to Newton AI before they start processing a user's request. Each prompt rule should ideally be a single concise sentence that gives Newton AI a very clear instruction. It usually helps to put important keywords like 'ALWAYS' and 'ONLY' in all capital letters. 'Model Prompt Rules' can be used in many ways, but they work best in use cases where you want Newton AI to catch certain request conditions and provide a certain response without Newton AI having to do all the usual work they would when answering a user's query, essentially overriding Newton AI's default behavior. This not only substantially lowers the cost of targeted requests, but also(and more importantly) substantially lowers the time it takes Newton AI to respond to those requests.
What kind of data can I use to train Newton AI? You can train Newton AI with virtually any web page, document, video, audio recording or plain text content. The easiest option is to add URLs or links to web pages that contain relevant information to the 'URLs' section of the create Newton AI model form. Additionally you can add URLs or links to entire sitemaps in the 'Sitemap URLs' form field. You can even train the Newton AI LLM with YouTube videos by leveraging the 'YouTube URLs' section of the create Newton AI model form. You also have the ability to upload documents, including PDFs, Microsoft Word documents and MP3s using the 'Additional Documents' section of the Newton AI model form. You can also create specialized Newton AI Training Modules on your Training Modules portal page that can be imported when creating a new Newton AI model.
What is a sitemap and how is it used to train Newton AI? A sitemap is a page or document that is typically hosted on the root of most websites, that contains a list of links or URLs to every page available on that domain as well as some additional metadata about those pages. When you provide a sitemap URL to the Newton AI LLM Trainer, the AI trainier fetches every page on the sitemap to use when training your Newton AI LLM model. If a sitemap contains URLs to pages that you don't want included when training your Newton AI LLM model, you can add those URLs to the 'Excluded URLs' field of the 'Create Newton AI Model' form on the Newton AI Models portal page. Additionally, you can learn more about sitemaps by reading this Google developer document.
What file types are supported when uploading additional training documents? The Newton AI trainer currently supports the following file extensions:
  • .pdf
  • .txt
  • .rtf
  • .html
  • .md
  • .docx
  • .json
  • .mp4
  • .mp3
  • .wav
How long does it take to train a Newton AI model? It can take anywhere from a few minutes to a few hours to train a new Newton AI model depending on current platform LLM trainer load and how much data is used to train your Newton AI LLM model. When you create a new model, the training of that model is put into a dynamically sized queue that prevents the platform from crashing under high load, and your model's status will be changed to 'QUEUED'. When it is your models turn in the queue, it's status will be changed to 'PROVISIONING', and the training job will be executed. When LLM training has been completed, the Newton AI model will be changed to a status of 'ACTIVE', and the previously active model's status will be changed to 'AVAILABLE'.
How should JSON(.json) files be formatted for training Newton AI? Training JSON should formatted as a JSON object with a single root property named "questions". The "questions" property is a collection of "question"/"answer" objects.

  "questions": [
        "question": "a question a user might ask",
        "answer": "the answer you would like the AI to give"
Are there any additional properties or information I can provide when using JSON to train Newton AI? Child "question"/"answer" objects can optionally include a "context" property, that is a string that contains contextual background imformation or prompting that Newton AI should consider when responding to similar questions, as well as an "priority" property, that is an integer greater than zero(1 is the higest priority) that defaults to 100 when not provided.

  "question": "a question a user might ask",
  "answer": "the answer you would like the AI to give",
  "context": "OPTIONAL - any background information",
   "priority": 100
Is there a tool or app that makes creating and editing training JSON easier? Yes, the Newton AI Training Modules portal page contains an editor that allows you to create, save, edit and download Newton AI Training JSON modules. Saved training modules can then be used to train Newton AI by selecting them on the 'Training Modules' of the create Newton AI form.
What are Newton AI 'Model Variables'? Newton AI Model Variables allow you to change selected values in provided training data in real-time(at any time) without having to re-train your Newton AI model. This is useful if you want to train Newton AI with URLs or other values that are subject to change. You can find your Newton AI Model Variables by clicking on the 'Model Variables' tab on your Newton AI Models portal page. To use a variable in your training data, simply replace the target value with variable name wrapped in curly braces({}}.

The URL for the support page is {supportUrl}.
Where is the best place to insert model variables? Newton AI Model Variables are primarily targeted for Newton AI Training Modules, but they should work in any training data you provide.
How does the Newton AI LLM trainer process audio and video files? First, the Newton AI LLM trainer seperates the audio from all given videos. Then the LLM Trainer transcribes all of the provided and processed audio into text transcripts. The resulting text transcripts are then used like any other text when training your Newton AI model.
Can I train Newton AI with YouTube videos? Yes, you can train Newton AI with YouTube videos by inserting links to relevant videos in the 'YouTube URLs' section of the create Newton AI model form. The selected videos will be downloaded and transcribed when your Newton AI LLM model is trained.
Are there any limits to how many Newton AI models I can train? Yes, we do place some limits on the training of Newton AI models in order to help ensure stability and availability of the Chicago AI Platform and its LLM trainer. LLM training rate limits are implemented on a per-API-Key basis. You can find your individual training rate limits by pressing the 'VIEW TRAINING LIMITS' button on your Newton AI Models portal page. If you would like to request a limit increase, or otherwise need support using our Newton AI LLM trainer, please contact Chicago AI and we will be very happy to help.

If the Q&A above doesn't answer all your questions, please don't hesitate to reach out and ask Chicago AI a question using the form on our Contact Us page.