Generative AI & The New Category of LLM Powered Applications

COBUS GREYLING
May 23, 2023
.
4 MIN READ

New approaches and tools built upon the strength of Generative AI are emerging to create engaging conversational experiences.

For years, I have been exploring the Conversational AI landscape, analysing how development frameworks are impacting the tooling used to accommodate the needs of a conversational interface.

As the development of chatbot frameworks has progressed, the importance of features such as natural language understanding (NLU), state machines for dialogue management, and knowledge bases has been highlighted, and additional functions have been included.

The progression of Conversational AI Framework functionality has been very linear, to say the least.

However, all of these developments were based on the existing framework (which I detail here). And again, it was largely a linear step progression built on existing Conversational AI framework principles.

Even the implementation of LLM functionality has lead to Conversational AI Frameworks becoming more alike in their attempts to differentiate. With LLM functionality complimenting the existing framework, without truly reinterpreting how the framework should look.

Hence I have embarked on a journey to explore the various generative AI tools and frameworks available for developing LLM-based conversational applications. I will be considering the framework of these applications in particular.

1️⃣ LLMs should not be used in isolation, as many organisations are currently doing. Creative and innovative ways should be explored to add value to the LLM API call.

2️⃣ The utilisation of LLMs alone is not enough to establish differentiation or a viable business model, nor can it guarantee the development of a highly efficient application.

And one of the reasons for this, is the general accessibility to LLMs and the general familiarity people of all walks of life now have of LLMs.

How do you build a competitive advantage when everyone is using the same models?

AI Is The Platform

There is huge potential for companies to disrupt by building applications from the ground up; making use of Large Language Model API’s. LLMs are ushering in a new paradigm of generative based applications.

There has been concerns of being at the behest of only a handful of LLM providers. There are continued efforts to open-source LLMs and language technologies.

LLM-based applications will require a framework which will need to run in some environment. Hence the emergence of frameworks like LangChain.

The Prompt Is The Program

Engineered prompts which invoke a LLM can be seen as a program which is compiled and run in real-time.

Applications will be required to engineer, manage (store, share, etc) and refine prompts.

Prompts can also be dynamic as with LangChain applications, where context, conversation memory and more is managed by compiling the prompt during the conversation..

Use-Cases

The use-case is of paramount importance, and the differentiator for start-ups.

Opportunities do exist, and need to be discovered in order to capture value. The greater the value added to LLM API calls, the more successful the start-up will be.

As seen in the image below, startups can own the UX and Prompt Engineering layers based a LLM.

The greatest opportunities are the areas of productivity, creative assistance and conversational experiences.

Chat Utilities

The market is moving away from the notion of a meta-bot, where a single overarching conversational interface serves as a single point of convergence.

There is a proliferation of smaller utility chatbot micro-frameworks. For instance FileChat.IO which is a tool to explore documents using artificial intelligence. Simply upload a PDF and start asking questions to your personalised chatbot.

Another example is a company called ChatBase which has a very similar product.

And another example, a company that lets you enter a URL, and summarises what the company’s url is all about.

Disposable Chatbots

I like to refer to it micro-chatbots, where chatbots are being treated as disposable utilities.

Chatbots are created for a specific task or instance, and can be discarded once used.

Content of a disposable chatbot can be a PDF, any document, a website, a conversation transcript between a customer and service representative.

Semantic search can be performed on text data in a conversational manner.

Frameworks Are Composable

New frameworks are approaching Conversational AI by combining foundation LLMs via a method of composability.

Composability is achieved with dialogs or conversations being seen as inter-relationship components. Various combinations can be selected and assembled for very specific user requirements.

LangChain is open-sourced software for the creation of LLM based multi-turn dialogs and managing contextual memory in a conversation.

Another example is for LangChain to be used to create a conversational QnA chatbot interface to the HuggingFace inference API.

Incumbent Technical Debt

Current Conversational AI Frameworks carry the burden of retrofitting their current development framework architecture with LLMs.

Seeking out valuable use-cases is a challenge working with the constraints of current systems.

New frameworks are being imagined and designed by innovators without any burden of existing technical debt or current framework constraints.

For instance, a company like DUST uses a completely different lingo than traditional chatbot frameworks. They speak about achieving a particular task. And chaining one or more prompted calls to models and external services (APIs or data sources).

I’m currently the Chief Evangelist @ HumanFirst. I explore & write about all things at the intersection of AI and language. Including NLU design, evaluation & optimisation. Data-centric prompt tuning & LLM observability, evaluation & fine-tuning.

Subscribe to HumanFirst Blog

Get the latest posts delivered right to your inbox