Skip to main content

Key Applications

Let's take a look at how this is happening in the real world, because understanding these applications will help you visualize the potential that exists in our own field.

Medicine​

Doctors face something similar when analyzing medical images. The difference is that instead of identifying structural problems in a building, they are looking for anomalies in the human body.

LLMs are being integrated with medical imaging technologies to help radiologists interpret scans with greater accuracy [15]. It's fascinating: these systems can analyze clinical reports written in natural language and correlate them with visual data, creating insights that increase diagnostic accuracy [16].

But it goes beyond that. Imagine having an assistant that reads all available medical literature and helps you make evidence-based clinical decisions. Systems like BERT and its variations have been specifically trained on medical texts, creating tools that can interpret clinical notes, understand test results, and provide well-founded recommendations [17]. It's like having instant access to the world's medical knowledge.

One application that particularly impresses me is the automation of medical documentation. Doctors spend hours filling out electronic health records—time they could be dedicating to patients. LLMs can transform natural conversations between a doctor and patient into structured records and automatically encode medical procedures [18].

Med-PaLM, developed by Google, represents the state of the art in this area. Using advanced prompting strategies—which are basically smart ways of "talking" to the AI—this model has achieved impressive results on medical benchmarks, reaching 67.6% accuracy on MedQA, an exam equivalent to the US Medical Licensing Examination [19].

Finance​

In the financial world, things get even more interesting. Think of finance as a sector that deals with an absurd amount of documents, data, and analysis—not unlike how we handle technical specifications, standards, and regulations in construction.

Trading algorithms now use LLMs to analyze news, financial reports, and even social media posts to predict market movements [20]. It's like having a super-analyst who can process thousands of sources of information simultaneously and identify patterns that influence stock prices.

For risk management, these systems interpret complex regulatory documents and identify potential compliance issues [21]. Just imagine: an assistant that reads all the changes in financial regulations and alerts you to what might affect your company.

BloombergGPT is a fascinating example of this specialization. Trained specifically with 363 billion tokens of financial data, this model can natively understand the jargon of the financial market [22]. When tested against other models on financial tasks, it consistently outperformed the competition, showing how much specialization matters.

Education​

In education, LLMs are creating personalized learning experiences in a way we've never seen before. It's like having a private tutor available 24/7 who adapts to your pace and learning style.

These systems can evaluate essays and academic papers, providing detailed feedback on strengths and areas for improvement [23]. Even more impressively, they can detect plagiarism and even create personalized exercises based on a student's knowledge level.

For us, who frequently need to stay updated with new technologies and standards, imagine having an assistant that can explain complex concepts simply, answer specific questions, and even simulate project scenarios for practice [24].

In law, a traditionally conservative field, LLMs are proving their worth in tasks that require meticulous document analysis—something that has much in common with the analysis of contracts and specifications we perform in projects.

Systems like GPT-3 are being tested on statutory reasoning, a fundamental skill in legal practice. Although they still have limitations, they can already interpret laws and apply them to specific situations with increasing accuracy [25]. GPT-4 even achieved a score in the top 10% on simulations of the bar exam [26].

Scientific Research​

Perhaps one of the most exciting applications is in scientific research. LLMs can assist in all stages of the research process: from literature review to hypothesis generation, data analysis, and even writing articles [27].

Imagine having an assistant that can read thousands of scientific papers and extract relevant insights for your research in a matter of minutes. Or one that can generate innovative hypotheses by combining knowledge from different fields [28]. It's like having access to an infinite idea laboratory.

Connecting the Dots​

Let's pause for a moment to absorb this. What do all these applications have in common? They deal with natural language—the way we humans naturally communicate. And here's the crucial point: for these systems to do their job, someone needs to know how to communicate with them effectively.

This is where a skill that is becoming fundamental comes into play: knowing how to create effective prompts. Because, in the end, the difference between a mediocre result and an extraordinary one often lies in how you formulate your question or instruction to the AI.

This leads me to the next step of our journey: understanding exactly what these prompts are and how they function as the bridge between your ideas and the power of LLMs.

References Cited in This Section​

[15] Zhi Li, Qiang Zhang, Qi Dou, et al. "A survey on deep learning in medical image analysis". In: Medical image analysis 67 (2021), p. 101813.

[16] Jun Zhang et al. "Medical image analysis with artificial intelligence". In: IEEE Transactions on Biomedical Engineering 68.5 (2021), pp. 1375-1379.

[17] Emily Alsentzer et al. "Publicly available clinical BERT embeddings". In: arXiv preprint arXiv:1904.03323 (2019).

[18] Benjamin Shickel et al. "Deep EHR: A survey of recent advances in deep learning techniques for electronic health record (EHR) analysis". In: IEEE journal of biomedical and health informatics 22.5 (2018), pp. 1589–1604.

[19] Karan Singhal et al. "Large language models encode clinical knowledge". In: arXiv preprint arXiv:2212.13138 (2022).

[20] Hans Buehler et al. "Deep learning and algorithmic trading". In: Financial Markets and Portfolio Management 32.3 (2018), pp. 239-260.

[21] Jin Li, Scott Spangler, and Yue Yu. "Natural language processing in risk management and compliance". In: Journal of Risk Management in Financial Institutions 13.2 (2020), pp. 158-175.

[22] Shijie Wu et al. BloombergGPT: A Large Language Model for Finance. 2023. arXiv: 2303.17564 [cs.LG].

[23] T. Susnjak. "ChatGPT: The end of online exam integrity?" In: CoRR abs/2212.09292 (2022). URL: https://arxiv.org/abs/2212.09292.

[24] K. Malinka et al. "On the educational impact of ChatGPT: Is artificial intelligence ready to obtain a university degree?" In: CoRR abs/2303.11146 (2023). URL: https://arxiv.org/abs/2303.11146.

[25] Andrew Blair-Stanek, Nils Holzenberger, and Benjamin Van Durme. "Can GPT-3 perform statutory reasoning?" In: CoRR abs/2302.06100 (2023). arXiv: 2302.06100. URL: https://arxiv.org/abs/2302.06100.

[26] Josh Achiam et al. GPT-4 Technical Report. 2024. arXiv: 2303.08774 [cs.CL].

[27] C. Zhang et al. "One small step for generative AI, one giant leap for AGI: A complete survey on ChatGPT in AIGC era". In: CoRR abs/2304.06488 (2023). arXiv: 2304.06488 [cs.AI]. URL: https://arxiv.org/abs/2304.06488.

[28] Yang Jeong Park et al. Can ChatGPT be used to generate scientific hypotheses? 2023. arXiv: 2304.12208 [cs.CL].URL: https://arxiv.org/abs/2304.12208.