Zehra Cataltepe is the CEO of TAZI.AI an adaptive, explainable Machine Learning platform. She has more than 100 papers and patents on ML.

The world of artificial intelligence (AI) is going through a huge transformation, fueled by the explosive growth of generative AI (GenAI) and large language models (LLMs). This revolution is reshaping the democratization, computational power and data requirements of LLMs and is also opening new ways of thinking in combining GenAI, AI and humans—working together as well as understanding and trusting each other.

The changing landscape also requires different solutioning approaches, challenging both deep and rigid point solutions as well as flat and oversimplified no-code solutions.

New LLMs

Especially for the midsize market, we are seeing huge opportunities. Open-source large language models (LLMs) are racing against OpenAI’s GPT and are winning. Fine-tuning the base models for specific use cases, especially the new Llama3 and Phi-3, may give results that are at least as good as, if not better than, using GPT-4. These lightweight models can perform quickly and well enough when they operate on consumer-grade GPUs, such as the one I have on my Mac.

Easing And Democratizing The AI Life Cycle

We are seeing the way AI workflows change thanks to generative AI. For example, for some use cases such as customer communications, rather than asking for lots of training data from our clients, we now ask for unlabeled data—which we label using LLMs. We determine the accuracy and reliability of these predicted labels using other LLMs and then ask for the domain experts to label some selected samples, provide the predicted label and reasoning and ask for feedback.

We can then use this “human-in-the-loop” feedback to continuously update the way LLMs operate by changing the prompts or fine-tuning the models. We have been using AI explanations, human in the loop and continuous learning for some time now, and being able to do this for GenAI as well is fascinating.

Composite AI And GenAI Together

We have decided to use both AI and GenAI in our solution library, sometimes within the same use case and sometimes in unique use cases across departments. For example, customer communications trending topics and sentiment solution uses GenAI to detect trending topics and AI to detect anomalies.

As a cross-departmental use case, VoC-Predictive-360 brings the predictions of AI solutions such as churn, upsell and cross-sell together with the GenAI-predicted topics for the current conversation. This way, agents can not only prioritize who to call next based on indicators like churn but also know the predictive “context” and respond to the customer in a hyper-customized way.

Trustworthy AI

The inclusion of human domain experts in the design, creation, approval, monitoring and updating of AI systems is becoming essential—especially with the introduction of powerful LLMs and rapidly developing regulations. We see the ability to provide human in command, human on the loop and human in the loop as well as the local open-source models as advantages in this regard.

It is an exciting time to be solving problems for and with your customers in cheaper, faster ways and bring more value to them.

Solution Approaches

When we meet customers today, most of the time, they have identified five to 10 use cases at the board or C-level. The usual approach of buying a point solution for each use case doesn’t work for AI because it is expensive and slow to buy, and it’s difficult to manage the change when there are many point solutions that use AI in different ways and with different integration and computational needs.

Moreover, AI-based decisions can sometimes even be too fast. You might prevent churn, but those prevented churners might actually be causing you to lose money. In the market, we are seeing GenAI wrappers that try to solve different use cases. These are mostly built on GPT models. However, we are also seeing these companies being wiped out with the next feature announcement from OpenAI.

There are no-code solutions entering into the huge market. These solutions demo well, but they don’t have the flexibility to handle the real-world use cases that need continuous monitoring and updates to AI. These approaches mostly lack continuous self-learning and human-in-the-loop approaches as well as the ability to quickly update the models, which enables handling the customer-facing teams’ need to deal with continuously evolving data and problems. No-code solutions mostly don’t use both AI and GenAI.

Hyperscalers such as Azure, Google and AWS are bringing their LLM-powered solutions to the market. Especially with mid-market financial services and insurance customers, Microsoft has been a trusted partner. Most customers also use Salesforce or Hubspot as their CRM. We believe these partnerships will be beneficial for the repeatability of use cases, especially when it comes to customer-facing use cases using GenAI.

Conclusion

The integration of AI and GenAI offers businesses unparalleled opportunities to streamline operations, gain deeper customer insights and drive innovation. However, navigating this complex and rapidly changing landscape requires a thoughtful solutioning strategy. Rather than chasing quick fixes or point solutions, organizations must prioritize solutions that combine the strengths of AI and GenAI, facilitate continuous learning and place human expertise at the heart of AI development and deployment.

By embracing a collaborative, human-centric approach, businesses can utilize the power of AI continuously and responsibly, ensuring that it serves their goals in an ethical and sustainable manner.

Zehra Cataltepe

Forbes Councils Member

Forbes Technology Council

Also Published on Forbes: https://www.forbes.com/sites/forbestechcouncil/2024/05/21/breaking-down-the-aigenai-landscape-and-solution-approaches/