Bigger languages ​​multiply: How SLMs can beat their bigger, more resourceful cousins

Two years after ChatGPT was released, the conversation about AI will not go away as companies in all industries look to use it. Major languages (LLMs) to change their business strategies. However, despite LLMs being powerful and reliable, many business and IT leaders have over-relied on them and overlooked their limitations. This is why I look forward to a future in which special-purpose languages, or SLMs, will play a large, supportive role in the IT industry.

SLMs are often referred to as “small languages” because they require less time and training time and are “more types of LLMs.” But I like the word “specialized” because it best describes the ability of these purpose-built solutions to perform specialized tasks more accurately, consistently and transparently than LLMs. By adding LLMs and SLMs, organizations can create solutions that take into account each individual’s strengths.

Believe in the LLM ‘black box’ problem

LLMs are very powerful, yet they are also known to sometimes “lose the plot,” or provide content that is distorted by their well-known studies and data overload. That habit is made more difficult because OpenAI’s ChatGPT and other LLMs are essentially “black boxes” that don’t show how they get the answer.

The inbox problem is going to be a big problem in the future, especially for companies and business-critical applications where accuracy, consistency and compliance are important. Consider healthcare, finance and legal services as good examples of professionals where wrong answers can have serious financial and life-or-death consequences. Regulatory agencies are already taking notice and are starting to demand it AI descriptorsespecially in industries that rely on data privacy and accuracy.

While businesses often use a “person-in-the-loop” approach to mitigate these risks, over-reliance on LLMs can lead to a false sense of security. Over time, carelessness can set in and mistakes can go unnoticed.

SLMs = high definition

Fortunately, SLMs are suited to overcome many of the shortcomings of LLMs. Rather than being designed for routine tasks, SLMs are designed with less focus and trained on specific devices. This helps them handle different languages ​​in areas where accuracy is critical. Instead of relying on large, diverse datasets, SLMs are trained on the data they are analyzing, giving them knowledge. moral wisdom to provide consistent, predictable and relevant responses.

This provides several advantages. First, they are clear, which makes it easier to understand the source and the reasons behind their results. This is especially important in highly regulated industries where decisions must be traced from source.

Second, their smaller size means they can perform faster than LLMs, which can be important for real-time applications. Third, SLMs give businesses control over data privacy and security, especially if deployed internally or built into the business.

Also, although SLMs may require specialized training, they reduce the risks associated with using third-party LLMs run by external providers. This control is especially important for applications that require consistent data management and traceability.

Focus on developing technology (and beware of vendors who overpromise)

I want to be clear LLMs and SLMs they don’t match. Instead, SLMs can augment LLMs, creating a hybrid solution where LLMs deliver more content and SLMs ensure accurate execution. It’s still early days even as far as LLMs are concerned, so I advise technology leaders to continue to explore the many opportunities and benefits of LLMs.

Additionally, while LLMs can handle a variety of problems, SLMs are not transferable to other areas. So it’s important to have advance knowledge of what you can use to deal with a problem.

It is also important for business and IT leaders to devote more time and attention to building the specialized skills needed to train, improve and test SLMs. Fortunately, there are many free resources and free courses available through popular sources such as Coursera, YouTube and Huggingface.co. Leaders should ensure that developers have enough time to learn and test SLMs as the AI ​​technology battle rages on.

I also advise leaders to watch friends carefully. I recently spoke with a company who asked me for my opinion on the claims of a technology company. All I could think was that either they were exaggerating their claims or they were just too deep in understanding the technology.

The company wisely took a step back and set up a proof-of-concept to test the sellers’ claims. As I thought, the solution was not ready for the big time, and the company was able to leave with little time and money invested.

Whether a company starts with a proof-of-concept or a live deployment, I advise them to start small, experiment often and build on early success. I have been working with small instructions and information, to get the results that are leaving when I feed the model more. That’s why slow and steady is the smart way.

In short, while LLMs continue to provide valuable opportunities, their weaknesses are becoming more apparent as businesses increase their reliance on AI. The combination with SLMs provides a way forward, especially in high-level fields that require precision and definition. By investing in SLMs, companies can future-proof their AI solutions, ensuring that their tools not only drive innovation but also meet the requirements for trust, reliability and control.

AJ Sunder is co-founder, CIO and CPO at Responsive.