Unit 21 Reading: Implementing AI tools in the Workplace

Dialogue: Implementing AI in the Workplace

SCENE: Two overseas business colleagues are in a Zoom call to discuss how to increase efficiency for common business tasks. The dialogue concerns Rachel, a British IT manager from London, and Alexander, an American software engineer from Rochester, New York.

Rachel: Good morning Alexander, and thanks for agreeing to meet on such short notice. Lately, I’ve been reading about how companies are using AI and LLMs to boost efficiency, and I’d like to see how we can apply it here. You’re our tech lead and resident AI expert, so tell me — where do we start?

Alexander: Good morning Rachel, glad to help. IT is all about automation and efficiency. Right now, a lot of manual work goes into analyzing market data, reviewing IT support tickets, and generating quarterly reports. Those tasks are repetitive and data-heavy, which makes them a great place to start.

Rachel: Aha, that makes perfect sense. Generating reports manually, is indeed time-consuming, and error-prone. So first off, what is the difference between AI and LLMs? Are they the same thing?

Alexander: Well, AI is just a general term, and an LLM is a specific piece of AI tech that is optimized for processing human language. The idea is nothing new, although recently there have been tremendous breakthroughs. The accuracy is getting pretty incredible.

Rachel: Understood, but I still worry about errors or unexpected results. We can’t afford distorted financial reports, or marketing data that misleads decision-makers.

Alexander: Uh, exactly. That’s why the foundation of any AI project is a clean, structured dataset. If our data isn’t consistent or reliable, even the most advanced model will produce distorted outcomes. Garbage in, garbage out, as they say.

Rachel: Right you are. Okay, so the first task is to clean and structure the data. Once we do that, what comes next?

Alexander: Uh, then we can start training our model on a curated dataset. I’m leaning toward machine learning, maybe even deep learning for the marketing side, since those systems can analyze customer behavior, and adapt as new data arrives.

Rachel: I like the sound of that. But how do these LLM systems actually learn?

Alexander: They learn mostly by example. During the AI training stage, we feed the new model large amounts of labeled data, and the algorithm adjusts its parameters to reduce errors. Later, during inference, it applies what it learned to new, unseen data. The better the training, the more accurate the inference.

Rachel: Okay. And does the AI keep learning after that?

Alexander: It can, through a process called fine-tuning. Once a model is trained on general data, we can fine-tune it using our company’s specific data — say, our historical sales or client communications. That makes the model more specialized for our needs.

Rachel: I’ve heard of GPT models, something about text generation. Would that be useful for us? And what does GPT even stand for?

Alexander: GPT stands for Generative Pre-trained Transformer. It's generative, meaning it creates new things. It's pre-trained, meaning its ready-to-go for most general tasks. And the real trick is the transformer, which means its neural network is really good at understanding and predicting common human speech patterns.

Rachel: Interesting. So GPT could handle marketing copy too?

Alexander: Exactly. That’s part of generative AI, which creates new content instead of just analyzing data. It can produce text, images, or even short videos. 

Rachel: But isn’t there a risk that the model just starts making things up? I recently read about these things hallucinating false information.

Alexander: Yes, that’s one of the biggest challenges. A hallucination happens when an AI produces something plausible but factually incorrect. That’s why we must build in validation checks, and use explicit quality-control rules to filter out unreliable content. The AI should assist, not replace, human review.

Rachel: Understood. Still, if we’re automating parts of finance or marketing people might feel threatened. How do we manage that?

Alexander: Through human intervention and clear communication. Automation doesn’t mean elimination — it’s about freeing staff from repetitive work. People can focus on creative and analytical tasks while a robust AI handles the dull routines quickly and accurately. And by setting transparent goals, employees will understand how AI supports them rather than replaces them.

Rachel: Right you are. What about communication across markets? Can AI help us with international coordination?

Alexander: Absolutely! Especially with natural language processing, or NLP. It allows machines to understand and generate human language. For instance, NLP could automatically translate client messages, summarize long reports, or even detect sentiment in customer feedback.

Rachel: That could be useful for global marketing. Could we also use it to handle customer queries?

Alexander: Definitely. We can design chatbots using prompts, the text instructions we feed to an AI model. If the prompts are clear, the answers will be more precise. But if they’re vague or ambiguous, the model might respond unpredictably.

Rachel: So clear prompting is as important as good data?

Alexander: Exactly. Even the best AI needs clear guidance. Well-written prompts ensure responses are explicit and useful.

Rachel: Right. Now, one last question. How do we measure success once we’ve deployed these tools?

Alexander: Well, we track performance metrics — speed, accuracy, and error rate — before and after intervention. If automation reduces time spent on reports or improves campaign targeting, that’s measurable progress. We’ll also gather employee feedback to make sure the tools are truly transparent and useful.

Rachel: Sounds good. Could you please prepare a document that summarizes and outlines all these proposed AI applications we just discussed? Then I will present it to the Board next week in Paris.

Alexander: Of course. I can't wait to hear what the Board thinks about all of this AI innovation we are proposing! I just hope it doesn't work too well, or else we may be putting our coworkers out of a job. Or even ourselves. 

Rachel: Well, I wouldn't worry too much! Someone still has to write the prompts and curate the datasets!

Alexander: True, and I think you are safe too, as someone still needs to double-check the reports and present them to the Board.

Narrator: End of Dialogue. Thanks for listening!

Oh my God, those guys are totally gonna get replaced by AI — and they don't even realize it yet. Anyway, I don't know about you guys… but I, for one, welcome our new AI overlords.