‘Large Language Models’: the art of knowing when and how
[Published in El Español]
Some weeks ago, I was lucky to attend an event on the future of AI in higher education in the United Kingdom. Different industry players discussed the appropriateness of its integration into the education model, although most of them admitted universities were still unmoved by the emergence of foundation models. They are not the only ones.
In Spain, according to a study by the National Observatory for Technology and Society, only 36% of companies have implemented AI technology in their business activities to gain greater knowledge from data. However, this trend is on the rise, as more and more organisations and public administrations are integrating AI-based solutions within their organisation.
In similar terms, one of the speakers at the event, an American businessman and owner of an international business school, told the audience that AI will be present in all areas of society in a few years, including, of course, higher education.
And the market seems to agree. A strong proof of this is that OpenAI has raised the largest funding round in history. It has raised $6.6 billion to continue developing new versions of its Large Language Models (LLM), GPT, as well as a corporate valuation of more than $155 billion.
Everything points to increasingly versatile and accurate foundation models, at the cost, of course, of a higher computational cost to execute the requests of the millions of users who use their conversational assistants on a daily basis. The debate is on: have the LLMs on which generative artificial intelligence solutions are based come to replace all of the above?
Commitment to LLMs, yes, but only when they are the best option.
During his talk, our American friend sketched a future scenario in which the ‘survival’ and success of human beings would be subject to those aspects that are impossible (in his opinion) for artificial intelligence to achieve: critical thinking, the capacity for reflection and, in particular, consciousness. Let’s say to ‘art’, rather than to ‘science.’
That is what I am appealing to from this platform, to “art.” Human beings’ ‘art’ of knowing when and how to use large language models as a response to a need, in any field, including business. Because we find that LLMs are now the algorithmic response we offer —also— for use cases such as classifying a customer’s comment, segmenting audience, or anticipating the demand for a service. It does not hold up.
Specialists have traditionally solved these and other challenges with other approaches: sometimes by coding human behaviour using rule-based systems. In other cases, with algorithmic techniques other than language models, such as machine learning models. The results are also very good, with a considerable decrease in the use of resources, both during the training of the model and during its lifetime.
Even the solution to problems related to understanding a text or making a summary, for instance, can be sometimes techniques other than LLMs such as Small Language Models (SLMs) which offer the necessary adaptability and accuracy at a much lower cost in terms of computing capacity, time and money.
In short, it is not that LLMs do not work. On the contrary, these large language models abstract very well capabilities related to classification and prediction, not just the generation of text, images and so on. It is a matter of deciding when they are the best option.
Surround yourself with a specialised partner to master “the art of knowing when and how.”
Going back to our UK event, our speaker appealed to the need for developing such critical thinking and other similar skills to become a successful professional in the future AI-flooded universe. And he recommended the business school he runs as a means to achieve this purpose.
It is very important to surround yourself with the right partners to successfully navigate this path. To decide which is the best analytical approach to each business problem. To identify the decision axes that lead us to choose one or another approach. And to choose LLMs when they are the best alternative, rather than because of inertia, a tendency or a lack of knowledge.
The race is hectic and has only just begun. It is essential to “surround yourself with a specialised partner to master the art of knowing when and how”. Otherwise, we may take risks, move away from efficiency, use a sledgehammer to crack a nut and, in short, give up a competitive advantage for which there is currently a large window of opportunity.
Now that sustainability is a goal that all humans pursue, let’s do it with “art”, with the “art” of humans. With the “art” of making efficient use of technological resources to achieve our goals, including making use of LLMs when this is the best option.
By: Juan Ignacio Moreno, Head of AI Solutions & Strategy at Innova tsn.
