Generative Intelligence #2: ChatGPT or Bard? How about domain-specific LLMs?

Generative Intelligence #2: ChatGPT or Bard? How about domain-specific LLMs?
Thanks, we'll send you periodic updates!
Oops! Something went wrong while submitting the form.

ChatGPT or Bard? How about domain-specific LLMs?

There's a palpable buzz in today's tech landscape, and it's all about Large Language Models (LLMs). These new AI tools are not just generating chatter; they're revolutionizing tasks, streamlining processes, and redefining efficiency.

Everyone's familiar with impressive LLMs like OpenAI ChatGPT and Google Bard. From drafting content that sounds oh-so-human to answering questions on any topic with the precision of an expert in the domain, they've certainly made their mark. But just when we thought we had the AI scene figured out and everyone was talking about choosing between ChatGPT and Bard, new specialized LLMs like Med-PaLM 2 and BloombergGPT are appearing. Tailored for healthcare and finance, respectively, they're showing us how industry-specific LLMs may have their say in the LLM landscape.

For many now, the problem will be choosing between breadth and depth when selecting an LLM. Do you embrace the versatile prowess of general LLMs, or do you opt for the laser-focused expertise of domain-specific models? Don't fret; we're here to guide you through this exciting terrain and help you make an informed choice.

General LLMs: Jack of all trades

General LLMs are digital know-it-alls. While they may not offer domain-specific advantages, they remain invaluable for a broad range of applications.

General LLMs, such as ChatGPT, provide insights across a wide spectrum of topics. Their adaptability means that whether you're exploring history, space, or finance, you'll receive a comprehensive overview. They are like AI digital polymaths.

The strength of these models lies in their ability to navigate across diverse inquiries with ease. With a ChatGPT at your side, multifaceted questions find answers and interdisciplinary discussions flourish, all while maintaining a balanced, holistic approach.

LLMs like ChatGPT don't just process information; they infuse it with creativity. Perfect for ventures that tap into imagination, from spinning stories to pondering intriguing scenarios, they blend vast general knowledge with a dash of (sometimes awkward) innovation.

General purpose LLMs like ChatGPT are Jack of all trades but also masters of none. They fall short on some domain-specific tasks.

Unveiling the might of specialized LLMs

As the brain thrives on information, specialized LLMs come alive with domain-rich data. BloombergGPT, for instance, was trained on a vast corpus of over 700 billion tokens, a mix of Bloomberg's curated financial documents spanning four decades and public datasets. The relationship is straightforward: the deeper and more nuanced the data, the sharper and more incisive the LLM’s analytical prowess.

BloombergGPT represents a turning point in domain-specific LLMs. Engineered to improve existing financial tasks like sentiment analysis, news classification, and question answering, it unlocks new opportunities for the vast quantities of data on the Bloomberg Terminal. The benefits of such pre-training can be game-changing.

BloombergGPT demonstrates how a domain-specific LLM can outperform other models of similar size on financial tasks, all while maintaining competitive performance on general-purpose benchmarks.

The great LLM conundrum: Breadth or Depth? How about both?

Navigating the realm of LLMs, we're met with the perpetual question: breadth or depth? Well, why not both? Imagine using a general-purpose LLM for a generalized domain overview and then diving deep with the domain-specific LLM to decode the nuanced intricacies of specific data.

Suppose you're in the asset management sector and need to formulate an investment strategy for an upcoming quarter. You could use a general-purpose LLM to gather broad market sentiments, emerging industries, and global economic indicators. This gives you a macro overview and highlights areas of potential interest.

Once you've zoned in on a specific sector, like tech or green energy, switch to a finance-specific LLM. Its finance-specific training allows it to analyze intricate financial statements, earning reports, and sectoral data to provide a more detailed analysis.

Implementing a solution with multiple LLMs

Rather than using different LLMs independently for these tasks, you can create applications that leverage several LLMs seamlessly for your workflows.

How would you implement a solution like this? Here are a few recommendations to help you get started:

  1. Create a unified interface: Instead of hopping between different platforms, create one that seamlessly integrates the insights from both the general and domain-specific LLMs using their APIs. Users do not need to know when different models are used to answer their queries.
  2. Synthesize the Data: While the generic LLM provides a panoramic view of market sentiments, ensure this information is a guiding framework for the domain-specific LLM’s in-depth analysis. The outputs of one LLM can be used as input for another LLM.
  3. Analyze iteratively: Don't analyze once and call it a day. As market conditions evolve, re-run your analyses, first getting an overview from the general LLM and then zooming in with the domain-specific LLM. This iterative process ensures your strategies remain aligned with both macro trends and sector-specific intricacies.
  4. Refine with Feedback: Integrate feedback loops where insights from one model can refine the queries posed to the other. For example, if the finance LLM identifies a sudden shift in a specific financial sector, loop back to the general purpose LLM to understand the broader market sentiment surrounding this shift.

In this symbiotic approach, you're not just making data-driven decisions but harnessing the best of both LLM worlds – wide-reaching comprehension and laser-focused expertise.

The future is not one LLM to rule them all.

What's most promising about this collaborative LLM approach is its growth potential. By leveraging their distinct strengths, these models optimize present tasks and set the stage for future advancements, promising a blend of flexibility and precision.

Despite all the talk of the next ChatGPT competitor and which LLM will dominate the market, at Locusive, we are convinced the future will be about creating workflow-specific solutions rather than using general LLM chatbots like ChatGPT for everything. Our solutions combine tools, data, and LLMs to create custom workflows that meet our client's needs. Be sure to check our free Locusive Slackbot to see what a custom solution that uses LLMs to search through your private data looks like!