2.1. Simple prompts
Let’s begin with something straightforward. I’ve compiled a selection of good prompts I’ve used with ChatGPT in the past 30 days. If you haven’t had the chance to interact with a large language model, this list should provide you with an overview of the fundamental capabilities of models like GPT-4.
“Can you provide 10 alternative names for this piece of code?” This could be for a method, variable, or constant.
“I’m not a native English speaker. Can you help me rewrite this sentence or paragraph to sound more like one?” Sometimes I’m just fine with sacrificing some of my personal voice for the sake of clearer and more professional English.
I often request translations, too. GPT-4 has already proven to be much more accurate and reliable than Google Translate in this aspect.
“Could you interpret this code for me? I’d appreciate a step-by-step explanation.” If the explanation is unclear, then it’s likely that my fellow engineers might also find it hard to understand, indicating a need for refactoring.
“Let’s discuss the new feature I’m developing. I’d like you to act as a team member and review my proposed implementation method step by step.” Alternatively: “I’m faced with a particular issue or feature to be developed. Could you suggest how you might approach it?” This is particularly helpful when I hit a creative roadblock.
Conventionality checks. ChatGPT typically defaults to a safe, conventional response. This can be limiting when looking for creative inputs, but in certain contexts, such as some engineering decisions, a conventional response might be ideal as it’s likely to be more widely understood.
Pros and cons. Ask ChatGPT for arguments in favor of and against a certain concept. The responses can help gauge how your thoughts align with or deviate from common viewpoints.
Finding synonyms. With ChatGPT readily available, it’s quicker to ask it for synonyms than to look them up in a dictionary.
Tackling anything I’m below average at. Take, for instance, naming characters in the short stories I write. I used to struggle with this, so now I ask ChatGPT for similar character names based on those I like. Even if none of the suggestions strike a chord, I can ask for the reasoning behind each name and use that as a springboard for further ideas.
“I need to write a performance review, create a blog article, or draft a long email, and so on—but I’m finding it hard to start. Can you ask me questions about the person or topic until we’ve gathered enough information to form a comprehensive review?” This approach helps overcome writer’s block, as conversations with ChatGPT are informal and free-flowing.
We can identify some basic patterns in these prompts. As a programmer, I often use ChatGPT like a peer to help solve coding problems. I also write a lot, so it serves as a writing partner for me. This approach reflects how many others interact with these models and the trend in early modern AI products. They were job-focused, leading to the emergence of AI roles like programmers, lawyers, security consultants, writers, and more.
Case study: Creating useful products with simple prompts
A year and a half later, after the hype subsided just a bit, I think of large language models as on-demand intelligence accessible through APIs. Whenever my software needs to use fuzzy logic or analyze unstructured data, I can just send an HTTP request and get the insights I need.
Let’s consider a real-world scenario. Imagine an applicant tracking system that filters candidates using specific keywords. Each resume must be reviewed by a recruiter to see if it matches their criteria. For example, they might be searching for a candidate with seven years of experience in a particular technology.
The top companies often get hundreds—if not thousands—of applications, resulting in a lot of repetitive manual work. Moreover, recruiters are usually not experts in technology. To compensate, they come up with their own heuristics, which might lead false positives or false negatives.
With large language models, rather than depending on these filters, we can tap into AI’s intelligence by making an API call. (Shoutout to Patrick McKenzie on Twitter for the inspiration behind this prompt.)
Suppose you’re looking for someone with at least seven years of Python experience. A developer’s resume indicates they’ve been using Django since version 1.8. Explain why you’d decide to include or exclude this candidate.
…and get a response:
Include them. Django 1.8’s release in 2015 implies over 7 years of Python experience.
I wasn’t sure myself since I’m a Ruby developer. But after looking into it, I confirmed this is the right decision. Great!
Building an applicant tracking system like that without AI would require engineers to either integrate a complex expert system directly into their product, or allow recruiters to set up a rigid job application workflow on their own. But LLMs can easily read the job listing and the resume by using their extensive knowledge—on demand.
AI bias in recruitment
A friend reached out after reading an earlier draft of this chapter and suggested that HR might not be the best example. That's true! Research indicates that built-in biases can influence candidate evaluations made by AI.
However, my point is that automated filters should be straightforward enough to minimize this issue. We don’t want the LLM to evaluate the entire profile and make any decisions on its own; it should just process the unstructured text, extract metadata at scale, and make very simple filters based on your existing criteria.
So here’s the best way to understand how these models are useful today, despite not yet being smarter than humans: if you assign them specific, concrete tasks, they can help you achieve good results efficiently. Turns out, my perspective on AI systems hasn’t changed much over the past decade: they can deliver what a thousand interns would, but much faster and at a lower cost.
The major difference this time is that the tech stack has become much simpler to use due to commoditization. What used to be accessible only to Big Tech only has now become available to everyone.