BASE / Content Hub / A world champion at your fingertips: the promise of LLMs

A world champion at your fingertips: the promise of LLMs

If the World Memorization Championships are the Olympics of human memory, then Large Language Models are the gold medallists in the digital realm.

Now, imagine if we could all harness the talents of these memory champions, feed them the content we need, and have them at our disposal 24/7. That’s exactly what Large Language Models (LLMs) offer.

In layman’s terms, an LLM is a system that can store enormous amounts of digital data while not missing a single detail. These models understand the intricate relationships between words and can generate reliable, human-like content based on what they’ve learned. 

From a slightly more technical perspective, as explained by Nvidia, an LLM is a sophisticated form of AI, trained on a massive corpus of text. To be more specific, LLMs are a unique class of deep learning architectures that are engineered to understand the interplay between context and meaning in sequential data like sentences. 

In this article we’re diving into the advancements in LLMs and why they’re a game-changer for the life sciences industry. While we won’t delve into the technical minutiae, grasping their transformative potential is crucial. 

Where do LLMs come from? 

As we pointed out in NNL: the future of data management in life sciences, the concept of Natural Language Processing (NLP) isn’t new. However, the notion of LLMs has only recently come into the limelight, especially in the eight months following the launch of ChatGPT. 

This technology has seen rapid advancements since its initial conceptualization around 2010. Some key milestones that have laid the groundwork include: 

2013 - Word2Vec

This paper represented a significant milestone. The focus of the article is representing words in a continuous vector space, allowing for the use of simple mathematics to compute analogies between words. This is a fundamental technique used in LLM methods to retrieve the most pertinent content to answer a question, for example. The paper also explores how to reduce computational cost and complexity for large corpora.

2014 - Sequence to Sequence Learning with Neural Networks

Recurrent Neural Networks (RNNs) were widely applied as language models. However, these models suffered from the long-term dependency problem, i.e., difficulty maintaining relationships between distant words. This paper proposes a solution to this issue by coupling two Long Short-Term Memory models, thereby increasing the reliability of RNNs for long sentences.

2017 - Transformer Architecture

This paper represents a shift in methodology. Previously, RNNs were widely used in NLP but suffered from being time-consuming and having long-term dependency problems. The paper introduced the Transformer architecture and the Attention layer, which addressed the long-term dependency problem while increasing the speed and efficiency of the architecture.

2018 - BERT

This groundbreaking paper, published by Google, addressed the limitations of unidirectionally trained models (i.e., from start to end of the text), which did not effectively capture the context for complex tasks such as question answering. BERT, which stands for Bidirectional Encoder Representations from Transformers, overcame this problem.

2018 - GPT

The first major model proposed by OpenAI demonstrated significant gains across numerous tasks using a two-step approach: 1) unsupervised generative pre-training on a large corpus of text, followed by 2) task-specific discriminative fine-tuning.

2019 - GPT-2 and 2020 - GPT-3

These papers demonstrate that a model’s capabilities are intrinsically dependent on the volume of training data. The key concept is that these models can “learn” to be multi-task language models using an unsupervised approach (i.e., with non-labelled data). This implication paved the way for the creation of models like GPT-3.5 and GPT-4, which can quickly learn and adapt to new tasks by “understanding” the general structure and representation of language, due to their extensive training datasets.

Why LLMs stand out 

Large Language Models are incredibly powerful, and their applications are yet to be fully understood. The main reasons for the hype around these new models are:

Depth of knowledge

Unlike traditional AI models, LLMs are akin to digital encyclopaedias. Their extensive training enables them to understand context more effectively, making them invaluable for interpreting complex data.


LLMs can be fine-tuned for specific tasks, whether it's analysing patient feedback understanding research papers, or generating summaries of vast datasets, they're up for the challenge.

Integration with existing systems

Their adaptability allows for smooth integration into existing digital infrastructures, enhancing both efficiency and capabilities.

Incorporating LLMs can significantly amplify the efficiency and productivity of your organization. 

Where to find LLMs

The landscape has become increasingly competitive, with several key players making their mark. A few notable mentions include:

Hugging Face

Think of this as your friendly neighbourhood guide to the world of LLMs. Hugging Face offers tools that simplify the adoption of these models across various sectors, including life sciences.

Tech Giants

Google, Microsoft, and AWS each have their own proprietary LLMs, each with its own set of strengths.

LLMs at BASE life science 

At our company, we combine cutting-edge technology with deep industry expertise. We lead the way in incorporating innovative technologies to optimize business processes and meet our clients’ specific needs. Our approach is agnostic, allowing us to integrate LLMs seamlessly into our clients’ existing infrastructures.

Curious about Google’s BERT model? We’ve got you covered. Operating on a Microsoft infrastructure? No problem, we’re compatible. Prefer to steer clear of the major players in favour of open-source solutions? We use a combination of Haystack and Hugging Face, leveraging top-tier open-source projects to add value to your organization. 

One practical application of LLMs in the life sciences sector involves sifting through multiple documents to extract key information. With access to digitized data, we can configure an AI assistant powered by LLMs to offer: 

Question and Answer functionalities

Text generation capabilities

Text summarization

This method has been instrumental in enhancing the efficiency of our data migration projects.


The integration of life sciences and LLMs represents a revolutionary shift, and the sector is ripe for the digital transformation promised by the applications of LLMs. 

As we stand on the brink of this digital transformation, understanding and embracing LLMs can redefine our approach to research, patient care, and drug discovery. They can also boost your internal productivity and output. 

Whether you’re a seasoned professional in life sciences or a tech enthusiast considering a career pivot, the future is ripe with promise. 

Get in touch

If you’re as captivated by the transformative potential of life sciences as we are, why not reach out? We’re always eager to discuss how we can collaboratively harness the power of innovation to achieve your goals. 


Manfredi Miraula

Senior Data Engineer

Content Hub

Related content

Thank you for downloading our whitepapers

We have sent a download link to your mailbox