There is no need to kill an ant with a sledgehammer.
Artificial Intelligence, or AI, is all the rage these days. Just about anyone you know has at least played with tools like ChatGPT and been amazed at the results they get almost instantaneously. ChatGPT is an example of what’s known as a large language model or LLM. Other examples of LLMs are Google’s Sage and Microsoft’s Gemini. All these models are built on incredibly complex algorithms, massive computing power, and a power plant of electricity. They are capable of delivering answers and solutions to an incredible range of problems.
That’s because these LLMs are trained on enormous datasets that enable them to analyze and build connections, allowing them to answer questions in an eerily human-like way. These LLMs can even write articles or entire chapters of books with a single prompt while writing your organization’s HR handbook for you in seconds. It can feel like magic.
However, LLMs have downsides that can make them impractical for most entrepreneurial organizations to adapt. That’s why many fast-growing companies are now exploring small language models (SLMs) as an alternative to LLMs. Let me explain what SLMs have and why they might interest you.
Resource Intensive
One of the primary downsides of LLMs is their incredible resource intensity. One constraint with any LLM is the massive amount of data required for training. Most organizations lack access to that volume of data, making it difficult for them to take advantage of it, especially since privacy concerns and the cost of accessing data continue to rise.
The processing power behind these LLMs is also enormous. There’s a reason why the stock prices of companies like Nvidia, which manufacture the chips that power AI machines, have skyrocketed. Building a data center capable of powering an LLM is costly.
Speaking of power, those AI data centers consume an incredible amount of electricity. Early on, companies established their centers near hydroelectric dams to maximize the use of low-cost power. Now, companies like Microsoft are upping the ante by signing contracts with companies that are restarting nuclear reactors that were formerly shut down. Who would have imagined Three Mile Island opening again? Why? Nuclear power is a clean, carbon-free energy source.
We’re just seeing the tip of the iceberg regarding other changes that will help accommodate our AI future.
An Alternative to Consider
Aside from the enormous financial and resource costs associated with powering LLMs, there’s another reason they might not be a good fit for entrepreneurial organizations: they’re overkill.
Because LLMs like ChatGPT are designed to answer any question or problem, they are, by definition, somewhat generalized. But you probably don’t need that ability if you’re running a fast-growing business.
That’s where SLMs are emerging as alternatives to LLMs. While they are far less sophisticated, SLMs can be taught to do smaller and more focused tasks to help augment the work done by your employees.
One example might be using an SLM to automate paying your vendors. When an invoice is sent in, the SLM can search for the original purchase order, cross-reference it with the vendor, confirm that everything matches, and then cut a check or initiate an electronic funds transfer—all in a matter of seconds.
What gives an SLM an edge in this work over, say, an automated script you could program is that the SLM can adjust to vagaries that arise in the process—such as an outdated vendor address—and then adjust accordingly, for example, by sending a confirmation email before sending payment. Using an LLM for this kind of job would be overkill, just like using one of your employees’ time. You don’t want to hunt an ant with a sledgehammer.
The point is that this isn’t taking away anyone’s job—it’s freeing up your employees’ time to do higher-value work. That’s human augmentation.
Another example is that you can use an SLM to aid in your hiring process by scoring candidates’ resumes and even their video interviews. Using an SLM could help you winnow down a list of candidates in a fraction of the time it would take your HR person or team. Again, not to replace the work done by your employees but to augment and enhance what they do with their time.
There is ample research that demonstrates the combination of AI and human oversight and input enhances the quality of technical work. Some studies show a gain of as much as 25% in accuracy.
Of course, anytime you rely on an algorithm, you need to be on alert for any unintentional bias or errors that could result. Whenever you use AI, it’s essential to adopt the mantra of ‘trust but verify’.
The Future of AI
AI will continue to play a significant role in how we conduct business in the future. The key is to find the application that works best within your company. While playing with LLMs like ChatGPT can save you time on specific projects, you should also consider how adopting more roles for SLMs within your business can become a valuable productivity partner for your employees. As my English friends are fond of saying, “horses for courses”. Pick the right tool for the job.
