Feature
Large language models, the “traditional” foundation of many AI solutions, are big, complicated to train and expensive to build. This combination has led some experts to believe that, for all the hype, the development of LLMs holds little promise as a business. That’s not necessarily bad news for specialized functions like HR.
The truth is, not everyone needs the full capabilities of a traditional LLM that can catalog and analyze great numbers of documents and answer queries on a wide variety of subjects. Smaller models can serve many needs just fine — and they’re cheaper, simpler to maintain and lend themselves to working with specific subject matter.
“The current one-size-fits-all model of generative AI has led to generic outputs, poor integrations, hallucinations and vulnerabilities,” Yellow.ai CEO Raghu Ravinutala told PYMNTS. “Companies seeking to integrate generative AI require technology tailored to their distinct needs, industry vocabulary and unique character.”
So, what does that mean for your organization?
Small Language Models Are Cheaper and Faster
Since the introduction of ChatGPT in 2022, discussions on AI have centered around LLMs. Often depicted as the engines of AI, they use vast amounts of data to answer questions or perform basic tasks.
In business, they’re commonly the foundations of chatbots and virtual assistants, answering questions for users, generating text or code, drafting summaries and translating languages.
However, they’re expensive. To run, they require huge amounts of computer resources as well as advanced hardware. They’re also prone to hallucinations, where they concoct inaccurate or misleading information and intermingle it with correct material, leading to confusion or outright business mistakes.
The fallout has a direct impact on the needs of users who require cheaper options. Because whatever the economic challenges, the demand for AI’s capabilities isn’t going away.
So, developers have increasingly taken advantage of a less-expensive alternative: small language models.
SLMs are cheaper to develop and maintain. They’re trained using information specific to particular areas, such as payroll or performance management. Their smaller scale makes them more manageable and accurate than LLMs, and thus more economical. They require markedly fewer resources to run and operate more quickly.
In short, they can underpin AI solutions dedicated to discrete areas, like HR and talent acquisition.
AI, Tailored for HR
Small language models offer HR a number of capabilities needed for both workforce and employee management.
Candidate screening is one example. SLMs can quickly evaluate a large set of resumes to determine which candidates fit best with a particular role. This will help HR more efficiently onboard employees by answering questions, providing documents and keeping track of each new hire’s progress.
Going beyond speed and accuracy, there’s a lot to like about SLMs from both the organizational and technological point of view. For one thing, they can run on-premises and are easier to secure than models operating in a vendor’s cloud.
Their faster performance also allows them to more quickly complete tasks like parsing resumes or answering employee questions. And because they’re easier to work with, they can be tweaked to minimize bias.
Their reliance on narrower, more focused data allows them to more efficiently keep up with regulatory changes and monitor compliance across a number of jurisdictions.
SLMs can also quickly analyze feedback, allowing employers to have a more up-to-date view of employee sentiment and the impact of employee experience. And, they can make learning more personalized by tracking each employee’s performance, learning preferences and skills needs.
The Small Language Model Economic Advantage
Gary Marcus, a New York University professor emeritus of psychology and neural science, is one of the prominent figures who believes the economics surrounding today’s LLM-focused approach to AI “are likely to be grim.”
To date, AI developers have believed the way to improve the technology’s capabilities was to continually expand or “scale” LLMs. The bigger the model, they said, the more advanced the solution.
From a business point of view, however, that idea is “just a fantasy,” Marcus wrote. While he doesn’t expect LLMs to disappear, “the economics will likely never make sense: additional training is expensive, the more scaling, the more costly.”
Eventually, Marcus believes, LLMs will become something of a commodity, without the ability to make back the billions of dollars that have been poured into their development.
“When everyone realizes this,” he said, “the financial bubble may burst quickly.”
Conference
Sep
16
HR Tech Conference Las Vegas 2025
Register
Conference
Oct
27
Gartner HR Symposium/Xpo Orlando 2025
Register
Editor's Note: Read about other topics related to the use of AI in HR:
Rethink Your HR Strategy to Include AI Agents — As AI agents make their way into our workforce, HR is called to adapt its workforce strategy.
Is AI Good or Bad for Recruiters? It's Complicated — AI can be of great support to recruiters, but applicants are now also using the technology to improve their chances. Some recruiters say that’s a problem.
One Place AI Can Help With Performance Reviews: Data Collection — With a growing number of use cases for AI, we looked into how the technology can help streamline a time-consuming part of the performance review process.
