Over the last year the tech world has been abuzz with excitement about the resounding potential of Generative AI. Conversations of which Faculty, as both experts and enthusiasts of GenAI, partners of OpenAI and Europe’s leading applied AI consultancy, have actively pursued and participated in. We have worked across public sector organisations from national security through to central government and the NHS.
Amid this age of AI, where technology is evolving at an unprecedented pace, we have noticed both excitement but also an equal measure of apprehension amongst individuals and organisations when trying to harness the power of these algorithms. As we see it, organisations face three main challenges.
The first, trying to identify an appropriate GenAI use-case, that not only helps their business, but is best suited to fulfil value and safety requirements. The second, creating a viable prototype and the third, battling between the want to experiment amongst the persistent concern of getting it wrong.
While the potential opportunities of LLMs are irrefutable, their implementation requires effective and careful planning to ensure it meets the needs of the business. Nevertheless, maybe the largest challenge is the breadth of choice and capabilities GenAI unlocks. It is undeniable that LLMs are a testament to the incredible strides made in AI research. They can generate human-like text, answer customer queries, summarise complex policies and even identify cybersecurity threats in vast textual data. Yet, these extraordinary capabilities have raised fundamental questions about their role in a rapidly changing world.
While you may regard LLMs with some trepidation, we propose a different perspective - one which encourages embracing and experimenting with the technology to better understand how to unlock the opportunities and gain valuable lessons about these cutting-edge algorithms. By confronting and exploring the digital world of LLMs, you can begin to demystify their workings and unlock invaluable lessons concerning their utilisation.
At Faculty we have expertise in the deployment and management of these algorithms to achieve real-world impact across sectors. We are already helping the Department of Education and other parts of government implement GenAI to solve real world challenges. We follow an incremental approach, beginning with small, secure steps and construction a foundation for team learning along the way.
Our work includes an eight-week study we ran alongside the Department of Education, where we harnessed chatbot capabilities offered by our partner, OpenAI, to assess draft local skills training plans submitted for review. Our goal was to succinctly summarise and compare primary insights and themes provided by local employers regarding training programs offered at nearby further education colleges.
Take, for example, our work with a large government client to consolidate and summarise an extensive collection of Human Resource rule documentation. Through the ideation, deployment and training of an LLM, we successfully improved the operational efficiency of the HR team in handling enquiries that had previously consumed a significant amount of time and effort sifting through this documentation. A task that was characterised by inconsistencies and a lack of uniformity. This challenge had been raised by numerous members of HR, emphasising its significance.
So what actually are Large Language Models (LLMs)? LLMs are deep learning algorithms that are trained on vast amounts of data (mostly from the online public domain) and consist of billions of tuned parameters. These parameters allow the sophisticated algorithms to generate text based on a probabilistic distribution - essentially they predict the most likely next word in a sentence.
Today, we have reached an ‘inflection point’. Today, GenAI typically outpaces human capabilities for use-cases where speed and volume matter more than the accuracy of content produced - models can effectively analyse huge quantities of data, but may miss some nuances. Take, for example, day-to-day activities such as reading comprehension, language understanding, and image recognition. With this in mind, in an age where transparency and informed decision making are so important, especially within government, the fulfilment of these requirements is also coupled with battling a seemingly unmanageable amount of data or documentation. LLMs can foster efficiency in the face of these complexities, all of which are critical building blocks for a more responsive and empowered workforce.
Nevertheless, while the realm of LLMs is exciting, it is essential to acknowledge the threats and reliability of these model outputs, which with hasty implementation can increase your organisation’s exposure to threats and inefficiencies. For example, biased algorithms, hallucination, legal uncertainty, and misinformation due to a lack of real-time data updates, effectively rendering the model "frozen in time".
Navigating the potential risks associated with threats stemming from LLMs can understandably be a daunting endeavour, especially if you are not familiar with using these models. It’s important to first consider three essential rules when applying the use of LLMs to your organisation’s specific context. Firstly, finding the right problem. Identifying a use case that considers how well the details of your problem will map with how LLMs work is a great place to start. Secondly, being thoughtful of how you are going to integrate that LLM into the right part of your decision flows and building the correct technical infrastructure around it, to streamline processes and produce intelligent decision systems. Finally, AI safety needs to be considered at all stages from use-case discovery to LLM implementation, to assess and manage the well known risks of LLMs and thus detect threats.
While you can implement these measures, it is also important to note you are not alone. At Faculty, we offer seasoned professionals across a range of sectors, who can expertly guide you through the intricacies of this landscape and safeguard your organisation against potential pitfalls.
In conclusion, the opportunities offered by LLMs are boundless, and the choice is in your hands as to whether you are part of this technological evolution, providing efficiency, transparency and better insights for everyday decisioning and tasks. It really all begins with a simple step: just start. Whether this be exploring potential use-cases, creating prototypes or simply deciding you are done with hoping this technological evolution will blow over (because it probably won’t).
The value of LLMs to your business is waiting to be untapped. All of this to say, if you do decide to embark on this adventure, we are ready when you are.