Two years after the discharge of ChatGPT, conversations about AI are inevitable as corporations across industries look to leverage large language models (LLMs) to remodel their business processes. But as powerful and promising as LLMs are, many business and IT leaders have turn out to be overly reliant on them and overlooking their limitations. For this reason, I anticipate a future by which specialized language models (SLMs) will play a bigger, complementary role in enterprise IT.
SLMs are commonly known as “small language models” because they require less data and training time and are “more streamlined versions of LLMs.” But I prefer the word “specialized” since it higher captures the power of those purpose-built solutions to perform highly specialized work with greater accuracy, consistency and transparency than LLMs. By complementing LLMs with SLMs, corporations can create solutions that leverage the strengths of every model.
Trust and the LLM “black box” problem
LLMs are incredibly powerful, but are also known to sometimes “lose track” or produce results that deviate from course on account of their generalist training and vast amounts of information. This tendency is made much more problematic by the undeniable fact that OpenAI's ChatGPT and other LLMs are essentially “black boxes” that don't reveal how they arrive at a solution.
This black box problem will turn out to be an even bigger problem in the longer term, especially for enterprises and mission-critical applications where accuracy, consistency and compliance are of utmost importance. Think of healthcare, financial services, and law as prime examples of professions where inaccurate answers can have huge financial consequences and even life-and-death implications. Regulators are already paying attention to this and can likely start demanding explainable AI solutions, especially in industries that depend on privacy and accuracy.
While corporations often adopt a human-in-the-loop approach to mitigate these issues, an over-reliance on LLMs can result in a false sense of security. Over time, complacency can set in and mistakes can slip through undetected.
SLMs = higher explainability
Fortunately, SLMs are higher suited to beat a lot of the constraints of LLMs. Rather than being designed for general tasks, SLMs are developed with a narrower focus and trained on domain-specific data. This virtue allows them to handle differentiated language requirements in areas where precision is of utmost importance. Instead of counting on massive, heterogeneous data sets, SLMs are trained on targeted information, giving them the contextual intelligence to deliver more consistent, predictable and relevant answers.
This offers several benefits. First, they’re more explainable, making it easier to grasp the source and rationale of their results. This is critical in regulated industries where decisions have to be traced back to a source.
Second, on account of their smaller size, they are sometimes faster than LLMs, which is usually a crucial factor for real-time applications. Third, SLMs offer corporations more control over privacy and security, especially when deployed internally or designed specifically for the corporate.
Additionally, SLMs may require specialized training initially, but reduce the risks related to using third-party LLMs controlled by external providers. This control is invaluable in applications that require strict data processing and compliance.
Focus on developing expertise (and be wary of vendors who overpromise)
I would like to be clear that LLMs and SLMs are usually not mutually exclusive. In practice, SLMs can complement LLMs and create hybrid solutions where LLMs provide broader context and SLMs ensure precise execution. Even though it's still early days in relation to LLMs, I at all times advise technology leaders to further explore the various possibilities and advantages of LLMs.
Additionally, while LLMs scale well to quite a lot of problems, SLMs may not translate well to specific use cases. Therefore, it is crucial to have a transparent understanding upfront of which use cases ought to be addressed.
It can be vital that business and IT leaders devote more time and a focus to constructing the particular skills needed to coach, fine-tune, and test SLMs. Luckily, there may be loads of free information and training available through popular sources like Coursera, YouTube, and Huggingface.co. Executives should ensure their developers have ample time to learn and experiment with SLMs because the battle for AI expertise intensifies.
I also encourage leaders to fastidiously vet their partners. I recently spoke with an organization that asked me for my opinion on the claims made by a specific technology provider. My guess is that they either exaggerated their claims or were simply unable to grasp the capabilities of the technology.
The company properly took a step back and implemented a controlled proof of concept to check the seller's claims. As I suspected, the answer simply wasn't ready for prime time and the corporate was in a position to get away with spending relatively little money and time.
Whether an organization is starting with a proof of concept or a live deployment, I counsel them to start out small, test often, and construct on early successes. I even have personally experienced working with a small set of instructions and data, only to search out that once I then feed the model more information, the outcomes veer off target. For this reason, a slow and regular approach is a prudent approach.
In summary, while LLMs will proceed to supply increasingly priceless capabilities, their limitations have gotten more apparent as organizations turn out to be increasingly reliant on AI. The addition of SLMs provides a path forward, particularly in high-risk areas that require accuracy and explainability. By investing in SLMs, corporations can future-proof their AI strategies and be certain that their tools not only drive innovation, but additionally meet the demands for trust, reliability and control.
