FAQs: Fullpath Agentic Chat Explained

Fullpath's Agentic is unique in that it combines the wide breadth of knowledge Agentic Chat holds with Fullpath’s Customer Data and Experience Platform proprietary data layer. This allows the chat to provide highly specific dealership-relevant answers to hyper-specific inquiries. When you give Agentic Chat a prompt or a question, it generates text as a response. However, it's non-deterministic, which means that even if you ask the same question multiple times, you might get slightly different answers each time.


 

A Agentic Chat  is a type of conversational AI chatbot that is powered by a large language model (LLM). The primary difference between a Agentic Chat chatbot and a regular chatbot is in the way they generate responses to user input.

A regular chatbot typically uses a set of pre-defined rules or decision trees to generate responses based on specific keywords or patterns detected in user input. While these chatbots can be useful for answering simple and straightforward queries, they are limited in their ability to understand the nuances of human language and generate complex or contextualized responses.

In contrast, a Agentic Chat  uses a large language model to analyze the context and meaning behind user input and generate responses that are more human-like and nuanced. These chatbots can understand and respond to a wider range of queries including those that are complex multi-part queries, or queries that require some level of creativity or problem-solving.

Overall, Agentic Chat chatbots are designed to provide a more natural and engaging conversation experience for users and will often produce better answers.

A practical example would be a question about safety (“Does the new Blazer have a safety rating? And how does it compare to the Tahoe?”), or car seats (“How many car seats can I fit in the back?”) A typical chatbot is not capable of handling this.

 

The main difference between a Agentic Chat and a manned chat (also known as a human-assisted chatbot) is the level of human involvement in the conversation.

A Agentic Chat powered bot uses a large language model to generate responses to user input automatically. These bots can handle a wide range of queries and conversations with minimal human intervention, making them useful for scaling customer support, handling routine queries, and providing 24/7 availability.

In contrast, a manned chat involves a human operator who is responsible for responding to user queries and providing support. These chatbots are challenged by the fact that the human doing the chatting is almost always located offsite and is not familiar with a dealership's inventory. This leads to a poor speed of response and insufficient answers. Even when the chatbot is maintained by the dealer BDC, the BDC members are often not available to handle the chats due to other obligations, leading to a fallback to a call center or a bot.

 

The GPT in ChatGPT stands for "Generative Pre-trained Transformer." 

 

GPT is a type of language model that uses deep-learning techniques to generate human-like text. The model is pre-trained on a large corpus of text data and can then be fine-tuned on specific tasks, such as language translation, text completion, or text classification. The GPT models are part of a family of transformer-based language models developed by OpenAI and are known for their impressive performance in a wide range of natural language processing tasks.

 

A Large Language Model (LLM) is a type of artificial intelligence model designed to process and understand natural language. It uses deep-learning techniques to analyze large amounts of text data and learn the patterns, structures, and relationships within the language. Large language models can generate human-like responses to queries using complete sentences and paragraphs, or generate new text from a given prompt.

 

These models are called "large" because they are trained on massive amounts of data, sometimes up to billions of words, or even entire internet contents. These large datasets enable the models to learn a wide range of linguistic patterns and nuances, and as a result, they are highly versatile. They can perform well across various natural language processing tasks, such as language translation, text summarization, sentiment analysis, and more.

Some examples of popular large language models include GPT-3, BERT, and RoBERTa, which have achieved impressive results on many language-related tasks and have received widespread attention and adoption in both academia and industry.

 

Fullpath Agentic Chat will respond in Spanish if the Shopper types and begins to converse in Spanish. Any leads captured in the conversation will be sent into the CRM in English however the linked transcript will show the chat conversation so the interaction will be visible to the dealership employees who read and process the leads!

 

Agentic Chat (Generative Pre-trained Transformer) is a specific type of Large Language Model (LLM) architecture. In other words, GPT is a type of LLM.

LLM is a general term that refers to any language model that is capable of processing and understanding natural language. These models are typically based on deep learning techniques and are trained on large amounts of text data to learn the patterns, structures, and relationships within the language.

GPT is a specific implementation of an LLM that uses a Transformer-based architecture. The GPT models are known for their impressive performance on a wide range of natural language processing tasks, such as language translation, text completion, text classification, and more.

Therefore, while LLM is a more general term that encompasses any language model designed to process natural language, GPT is a specific type of LLM architecture that has gained widespread attention and adoption due to its impressive performance on many language-related tasks.

 

Yes, Fullpath Agentic Chat does this just like any chat conversion form.

 

Yes, Fullpath Agentic Chat transcripts are sent with the ADF and are available on the visitor's Shopper page in your Fullpath Dashboard.

 

From OpenAI's policy:

Starting on March 1, 2023, we are making two changes to our data usage and retention policies:

1. OpenAI will not use data submitted by customers via our API to train or improve our models, unless you explicitly decide to share your data with us for this purpose. You can opt-in to share data.
2. Any data sent through the API will be retained for abuse and misuse monitoring purposes for a maximum of 30 days, after which it will be deleted (unless otherwise required by law).
    

Implementation requires putting the Fullpath code snippet on your website. If you are already a Fullpath customer, you're ready to roll!

 

For those who have purchased Fullpath Agentic Chat, the interface will replace the native Website Engagement chatbot.

 

No, you only gain access to Fullpath Agentic Chat. To gain access to Website Engagement's functionalities, you need to sign up for a Website Engagement package.

 

It is not co-op eligible as of yet. We will provide updates as they become available!

 

That’s up to you and your OEM. We’re happy for you to use it!

 

No, it is covered by the existing Fullpath code snippet. If you do not have a Fullpath code snippet you will be prompted to add one during onboarding.

 

No, but the lead received in your ADF will contain the entire exchange, so you can pick up right where your shopper left off!

 

Reporting is the same as it is in Website Engagement.

 

Yes.

 

We do provide basic support and are happy to answer questions. We can and do provide the ability to create documents, or custom instructions, which Agentic Chat can use to guide the conversation or respond to certain questions or phrases. Customers may request specific instructions be applied through their CSM, who will open a ticket for Technical Support and escalate accordingly. Typically these instructions can be applied to the account within 48 hours (usually less).


 

Why yes, it does. Let’s define what we’re talking about here. Hallucination in AI refers to the generation of outputs that may sound plausible but are either factually incorrect or unrelated to the given context. These outputs emerge from the Open AI model's inherent biases, lack of real-world understanding, or (and in our case, most probably) from training data limitations around vehicle shopping, selection, and buying processes and interactions. In other words, the AI system "hallucinates" information that it has not been explicitly trained on, leading to sometimes unreliable or misleading responses. This problem is NOT unique to Agentic Chat, rather it's a problem all LLMs struggle with when trying to interact with users in a question and answer format.

There’s no easy way to fix this. We can, however, allow the system to learn and improve over time. Additionally, we can provide bumpers and guides for how the Agentic Chat can navigate specific scenarios or responses (see above regarding Custom Instructions).

A disclaimer to this effect is always present at the beginning of the chat interaction.

 

Like all GPTs, it is possible that Agentic Chat may, in some rare cases, provide awkward or inappropriate responses. Just like the humans you hire who may respond to custom inquiries inappropriately, you have to train your employees. At Fullpath we take several steps to mitigate this risk and train our Agentic Chat including: 1) We implement guardrails against such things in the system level prompt 2) We maintain a record of all chats so we can show nefarious prompting in the event that there is a claim  that the "shopper" was able to break these guard rails (this tool is pretty handy as it also serves to troubleshoot any other issues and resolve) and 3) we also include disclaimers in all chats disclosing that the shoppers are using an LLM GPT and none of the responses reflect the opinions of the Dealership or it's employees.

Lastly, our product and development teams are working on incremental changes weekly. We are using a method called prompt hardening which is assisting us in ensuring Fullpath Agentic Chat is more reliable and controlled in the outputs. The goal of prompt hardening is to mitigate potential risks associated with AI models by reducing their sensitivity to certain types of input or by constraining their behavior. Prompt hardening is an ongoing process that requires continuous monitoring and refinement to address emerging issues and concerns.