“Hi ChatGPT, write me a blog about using large language models“

When people think of artificial intelligence, they often first imagine tools like ChatGPT or Microsoft Copilot—examples of large language models (LLMs). LLMs are perhaps the most visible part of AI. They’ve advanced incredibly fast and are very versatile. They can be used for a multitude of tasks such as information retrieval, translation, writing, coding, or formatting clear additional details for an interpretation service order.

However, when writing prompts for a language model, it’s essential to be precise, as these models understand instructions in their literal meaning and have some limitations. This blog summarizes what large language models are and are not good for.

LLMs as workplace assistants

An LLM can be an excellent assistant for everyday work tasks. They excel at general information retrieval, data analysis, and summarization. For instance, you could ask a language model “How much does a one-hour remote interpretation session in Finnish-Arabic typically cost?” and it will likely provide you with an accurate price estimation. When it’s time to order such an interpretation service, you can ask the model to provide concise instructions for using the Youpret app, and it will handle this task, too, effectively.

Once your organization is familiar with interpretation services and using them regularly, a supervisor might request a report on the resources allocated to such services. A language model can help here too. By inputting information such as the dates, costs, durations, and languages involved in the interpretations, you can ask the model to create a clear graphic presentation, which provides a solid basis for drafting the report. When writing such a prompt, however, you must ensure no sensitive information that should not leave your organization is provided to the AI.

Read more about safe use of AI

In addition to retrieving and processing information, LLMs can effectively produce various types of content, such as emails, reports, meeting notes, and presentations. Thus, you could first use a model to create a graphic presentation to make writing the supervisor-requested report easier and then ask it to draft the entire document. It’s wise to instruct the model to utilize the groundwork it did as the foundation for the report, as doing so ensures the accuracy of the information covered.

Tasks like these might require a more advanced prompt. For instance, you should ask the model to adopt a relevant role: "You are a professional in financial management and analytics. Your task is to write a report on the use of interpretation services in our organization. Pay attention to the number of sessions, costs, durations, and interpreted languages. Use the graphic presentation you created as the basis for this report."

Language models are also great at extracting key points from lengthy documents and large datasets, such as meeting notes or customer feedback. For example, you could ask a model to summarize the meeting where the report regarding the use of interpretation services in your organization is discussed or extract critical points from customer feedback regarding how interpretation services have improved the customer experience in your organization

Shortcomings in precision and ethics

The weaknesses of LLMs emerge in tasks requiring particularly accurate information. They cannot be expected to have detailed industry-specific expertise. This lack of specific knowledge currently limits their usability in fields such as healthcare. When dealing with complex inputs like patient records or descriptions of medical conditions, a language model may “lose its train of thought,” resulting in off-topic or incorrect responses. Similar issues can be observed with machine translators trained on general knowledge.

Current LLMs also struggle with processing information that is not widely available. For this reason, detailed personal profiles may contain errors in dates, names, or other fine details.

As mentioned earlier, caution is necessary when handling sensitive data with language models. This is because users cannot fully know how these models process the provided data, how it might appear in other users’ responses, or how it is stored. Additionally, AI does not comprehend ethics as humans do, which means language models may handle personal data like any publicly available information, potentially leading to discrimination or breaches of privacy.

Read more about the accuracy and ethics of AI systems

---

If you asked ChatGPT, “How do I order a Youpret interpreter?” the response might look something like this:

"You can order a Youpret interpreter through their user-friendly app, available for Android and iOS devices. The app’s order form prompts you to input the necessary details for interpretation, such as the language pair, the interpretation topic, and whether it’s a scheduled or urgent session. Youpret also offers free training on how to use their app."

Book a free onboarding training

Author:

---

---