Stop Wasting Time on ChatGPT—Here’s How to Get the Answers You Need (Ideas Lab)

Introducing an NLP Solution to Get Straight to the Answers You Want

Intro

Imagine for a moment—you’ve scheduled vacation time at work and you don’t have anywhere to go! ChatGPT to the rescue! You open up a snazzy new chat window in your favourite LLM and pose the question, “Where should I go for a quick getaway?”. Instead of getting a nice, tailored response, you get the equivalent of a Google search result. Things that you might find in any generic, SEO flooded top 10 list of garbage. Is this what the millions of dollars of investment in OpenAI and NVIDIA computing power has amounted to? A glorified search engine with file attachment and image generation capability. If you’ve used ChatGPT, you’ve likely come across this issue—you just wanna casually “chat” and get some decent results. However, the name ChatGPT is deceptive in some ways, and it can take quite a few back-and-forths to get close to answers you were looking for.

What’s an idea? Is it simply a worthless cognitive process? Can ideas have value on their own? Is it even worth bothering with idea generation if it’s not immediately backed up with action? We’re going to find this out together! Since I was a wee lad, my brain’s been a place of thinking—as opposed to all those other non-thinking brains out there. I was able to make connections that others couldn’t. The trouble was I didn’t fully realise this until much later in life. Furthermore, it was only very recently that I had a big enough epiphany to push me into action, burst through the self-doubt, stop worrying about what other people think, and simply embrace this gift for what it is.

Problem Statement

ChatGPT can be very effective when given specific instructions, however, the prompts required to reach its full potential can be complicated for non-power users. This likely contributes to people's impressions of ChatGPT as it being "underwhelming". Most users, myself included, don't want to dedicate so long crafting the perfect prompt for every new chat window we open in ChatGPT. It is, after all called ChatGPT, not spend-time-creating-a-complex-perfect-prompt-to-reach-its-full-potential-GPT.

While the Language Learning Models become "smarter", they tend to be capped or limited when it comes to usability. User experience should be at the forefront of functionality. Yes, ChatGPT is indeed smart and we've really seen nothing like it previously. Does anyone remember IBM Watson? Watson was so smart it could play chess! Watson was a one-trick pony. Playing chess is something that can be achieved with enough maths. Other humanly endeavours are much trickier for computers to "understand" and fenagle their way through. Humans have nuance, we’re biological beings. It's much more difficult to replicate this with what amounts to an unimaginable collection of 1's and O's.

It isn’t so much that ChatGPT lacks “intelligence”, far from it—most users simply don’t know how to ask the right questions. This is also an issue in general for many, though that shouldn’t stop companies from crafting solutions with the average users themselves in mind. ChatGPT’s answers are reflective of the questions being asked, but it doesn’t have to be that way. Without a well-crafted, uber-engineered prompt, even the snazziest of snazzy, the super-de-buperiest of Large Language Models (LLMs) will give underwhelming responses. In this Ideas Lab article, we’re gonna explore a potential way in which to bridge the prompt engineering gap that exists within ChatGPT.

How can we make ChatGPT more accessible to the average user, by automating the complex task of prompt engineering?

Casualwear to the Prom

What is technically “wrong” with casual prompting—chatting, if you will, with ChatGPT? After all, when we’re chatting with friends, we don’t need to use complex, context-filled prompts before every conversation, do we? So why is this the case with ChatGPT? I thought it was smart!

What’s in a Prompt?

To better understand what this problem looks like in practice, we can take two test cases—one is how most people interact with ChatGPT, and the other is how people ideally interact with ChatGPT for the best results.

As I mentioned earlier, users expect to able to simply "chat" with ChatGPT—we expect the system to understand us, what we want. This is the implied ability. However, this is not the reality. I assume that many users (the people, like you and me) don't understand how much better their experience would be simply through effective prompting. I'm sure many of you reading this would have heard the term ”prompt engineering” at some point, though it seems to have died down along with the initial ChatGPT / LLM hype. We've passed the honeymoon phase. Prompt engineering is a way for us humans to provide enough context to an LLM (the computer) so that it can better understand our intent and give an answer that is most inline with our original query / question.

When we're having conversations with other humans this isn't as much of a problem because we pick up cues and context as we go along. For example, if I have a friend who is obsessed with cars, I might expect analogies from them to be car-related. When I propose an idea to them I can put it in the form of a car analogy because I more or less know they'll understand what I mean. It's not necessary for my friend to write out a detailed prompt for me every time we have a conversation. Computers and computer systems such as LLMs cannot pick up on social cues and it's not possible to adjust themselves on-the-fly to suit the slight, nuanced differences between various conversations and people. This is prompt engineering—a simplified way for us to give ChatGPT (LLMs in general) the necessary context for a given conversation.

Prompt Engineering Enters the Stage

Prompt engineering is giving specific, detailed instructions to an LLM so that you, the user, can get quality information in the way that you want it with minimal “back-and-forth”. When you open a new chat window, prompt engineering helps set the stage, set the context of how you want the conversation to go. Think of it as you introducing yourself each time you open a new chat—your preferences and so on. Not only that, but it would be like introducing yourself to someone and then telling the other person how you want them to behave.

Prompt Examples—the Engineered, the Casual (and the Ineffective)

Let us gaze upon some made-up, real world examples of prompts to get us warmed up and seeing the problem for what it is.

The Casual

Casual prompts are the ones I expect most people will be plugging into ChatGPT for the most part. They don’t have much specificity or context and are going to yield generic responses.

What’s a good dinner recipe?

How do I get a promotion at work?

Help me improve my public speaking skills.

What’s the best way to start a blog?

The above all have the following elements:

• Vague

• Too general

• Lacking context

Elements of a ‘good’ prompt:

• Clear

• Specific

• Structured

• Includes context

• Constraints

• Instructions for output format

Prompting Test Scenario

To actually test this and show exactly how much more effective prompt engineering is, let’s smash out our own scenario, with two test cases—casual prompting vs. prompt engineering.

Jimmy Jabrone, the IT homie

Jimmy is as laid back as they come. To everyone he appears like the most chill dude on the planet. The trouble is, he’s been internalising a lot of stress and it’s starting to impact his work. IT can be pretty stressful at times. For this scenario, Jimmy wants to find some actionable strategies for reducing stress in his IT job. This is a high-stress, tech-related job. The advice should be practical.

Goooooaaaaal

The overall goal is to get ChatGPT to give Jimmy a comprehensive list of practical, step-by-step strategies for reducing stress at work. As we mentioned, the advice should be tailored specifically for Jimmy in his high-stress, tech-related job.

How are we gonna decide when the goal is reached?

Glad you asked, my awesome reader!

Let’s set the criteria. This way, we can pretend like we’re all science-y and proper and all that kinda stuff.

Goal Criteria

ChatGPT’s response should cover the following:

• Detailed list of strategies—still a bit fluffy, a little ambiguous, but we’ll go with it for now

• Techniques for during work hours and after-work hours

• Tailored to someone working in a high-stress, tech-related job

• Practical (they can be incorporated into Jimmy’s existing routines without massive changes or time investment)

• Clear and easy to follow

• The response requires no further follow-up (the above points are covered)

Scenario Setup

Using two temporary chat sessions, one for each test case, I’ll run through each test case and show the journey to reaching the goal. I am choosing to use the temporary chat sessions because they don’t include my custom instructions—this ensures the testing is closer to what a real user might experience and it’s the default state for ChatGPT. The number of follow-up chats will be recorded for each until the goal is reached.

Below are the initial prompts:

Casual prompt

How can I reduce stress at work?

Engineered prompt

Act like a stress management coach specialising in high-stress environments within the tech industry. I need a comprehensive, step-by-step guide on reducing stress at work for someone working long hours in a tech-related job. Focus on strategies that can be implemented during the workday and after work hours. Please avoid generic advice like 'take breaks' and instead provide actionable, specific techniques.

The below video is the test in action. Test Case 01 is the “casual user” and Test Case 02 is the engineered prompt.

Results

Interestingly, and as you can see in the video above, it took the “casual” prompt test case 5 times longer to reach the equivalent answer, and that answer was actually slightly worse than the engineered prompt. You can be the judge of which answer is the best as there is a subjective element to it. Test Case 01 also used 4 prompts vs. 1 prompt for the engineered version. Keep in mind, this is biased toward the lower end, particularly for Test Case 01 because the goal was already known. In a realistic scenario, if the user has a lot of uncertainty about what they want, it would take a lot more than 4 prompts to feel ‘satisfied’ with the result, due to the subjective nature involved from the user’s perspective.

Below are the final answers from both test cases.

Casual Prompting Final Strategies:

Here's a detailed, practical, and actionable list of strategies that cover both work hours and after-work hours to help you reduce stress in your tech-related job:

During Work Hours:

1. Automate Repetitive Tasks

   • Action: Automate tasks like code deployment, testing, or generating reports using tools like Jenkins or custom scripts.

   • Implementation: Identify one repetitive task and automate it. Review the results after a week and refine the automation if necessary.

2. Use the Pomodoro Technique

   • Action: Implement the Pomodoro technique with 25-minute work sessions followed by 5-minute breaks.

   • Implementation: Set up a Pomodoro timer on your computer or phone and commit to completing four cycles before taking a longer break.

3. Batch Similar Tasks

   • Action: Group similar tasks like emails, code reviews, or meeting preparations, and handle them in one session.

   • Implementation: Schedule two 30-minute slots in your calendar each day to focus on these batched tasks.

4. Optimize Communication

   • Action: Encourage asynchronous communication to reduce interruptions. Schedule time for checking messages rather than responding immediately.

   • Implementation: Set "Do Not Disturb" hours in your messaging apps during deep work sessions and communicate your availability to your team.

5. Take Short, Mindful Breaks

   • Action: Use your breaks to practice mindfulness, do light stretching, or take a short walk.

   • Implementation: Set an alarm to remind you to take a 5-minute break every hour. Use this time to step away from your desk.

6. Refactor Code Regularly

   • Action: Allocate time in your sprint for code refactoring to manage technical debt.

   • Implementation: Schedule 1-2 hours per week dedicated to refactoring. Treat it as a priority task in your sprint.

7. Practice Mindful Debugging

   • Action: When stuck on a difficult bug, take a short break to clear your mind before continuing.

   • Implementation: Implement a rule where you step away from your desk for a 5-minute walk if you haven’t made progress on a bug in 30 minutes.

8. Set Clear Work Boundaries

   • Action: Define your work hours and stick to them. Avoid checking emails or work messages outside of these hours.

   • Implementation: Set an end-of-day ritual (e.g., shutting down your computer) to signal the end of your workday. Communicate your boundaries to your team.

9. Use Ergonomic Tools

   • Action: Set up an ergonomic workspace to reduce physical strain during long coding sessions.

   • Implementation: Invest in an ergonomic chair, adjustable desk, and monitor stands. Adjust your setup to maintain proper posture.

10. Document Daily Progress

    • Action: Keep a log of your daily achievements and challenges to track progress and reflect on improvements.

    • Implementation: Spend 5 minutes at the end of each workday documenting what you’ve accomplished. Review this weekly to assess progress.

After-Work Hours:

11. Unplug and Disconnect

    • Action: Create a clear separation between work and personal life by unplugging from work devices after hours.

    • Implementation: Turn off work-related notifications on your phone after your set work hours. Use this time for personal activities or relaxation.

12. Engage in Physical Activity

    • Action: Incorporate regular physical activity into your routine to relieve stress and boost energy.

    • Implementation: Schedule at least 30 minutes of exercise, like walking, running, or yoga, after work. Commit to doing this at least three times a week.

13. Pursue Hobbies and Interests

    • Action: Dedicate time to non-work-related hobbies or interests that bring you joy and relaxation.

    • Implementation: Set aside at least one hour each evening for a hobby you enjoy, such as reading, cooking, or playing an instrument.

14. Practice Mindfulness or Meditation

    • Action: Engage in mindfulness practices or meditation to unwind and manage stress after work.

    • Implementation: Spend 10-15 minutes each evening practicing mindfulness or meditation using an app like Headspace or Calm.

15. Plan for Tomorrow

    • Action: Prepare for the next workday by organizing your tasks and setting priorities before bed.

    • Implementation: Spend 10 minutes each evening reviewing your task list and prioritizing for the next day. This helps clear your mind for better sleep.

16. Stay Socially Connected

    • Action: Maintain a healthy work-life balance by staying connected with friends and family.

    • Implementation: Schedule regular social activities, even if it’s a virtual hangout, to decompress and maintain strong relationships.

17. Maintain a Healthy Sleep Routine

    • Action: Prioritize sleep to ensure you're well-rested and ready for the next day's challenges.

    • Implementation: Establish a consistent bedtime routine, aiming for 7-8 hours of sleep each night. Avoid screens for at least 30 minutes before bed.

18. Engage in Continuous Learning

    • Action: Dedicate time to learning new skills or technologies, but do so at a pace that doesn't overwhelm you.

    • Implementation: Set aside 1-2 hours each week for professional development. Focus on one topic at a time to avoid information overload.

19. Reflect on the Day

    • Action: Take a few minutes each evening to reflect on what went well and what could be improved.

    • Implementation: Write a short journal entry or make mental notes about your day. Identify one positive takeaway to end the day on a good note.

20. Take Micro-Vacations

    • Action: Plan short getaways or "staycations" to disconnect from work and recharge.

    • Implementation: Use your PTO for a long weekend every few months. Fully disconnect from work during these breaks to maximize relaxation.

By integrating these strategies into both your workday and personal life, you can create a balanced approach to managing stress. This holistic approach not only improves your productivity and focus at work but also enhances your overall well-being and quality of life.

---

Prompt Engineered Final Strategies:

Comprehensive Stress Management Guide for High-Stress Tech Environments

In high-stress tech environments, where long hours and demanding projects are the norm, managing stress effectively is crucial for both mental well-being and productivity. Here’s a step-by-step guide with specific, actionable techniques that can be implemented during the workday and after hours.

During the Workday

1. Morning Mental Preparation (10-15 minutes)

   • Mindful Transition: Before starting work, take 5 minutes to transition from home mode to work mode. Sit quietly, breathe deeply, and visualize your day, focusing on how you want to handle challenges.

   • Priority Mapping: Spend 5-10 minutes mapping out your top 3 priorities for the day. Write them down on a sticky note and place it somewhere visible. This keeps you focused on what truly matters and reduces overwhelm.

2. Pomodoro Sprint Optimization

   • 90-Minute Work Cycles: Implement work cycles of 90 minutes followed by a 15-minute active break. This is more effective than shorter cycles as it allows for deeper focus without the stress of constant interruptions.

   • Micro-Stretching Routine: During these breaks, perform a specific stretching routine that targets tech-related tension areas like the neck, shoulders, and wrists. This helps prevent the build-up of physical stress.

3. Environmental Calibration

   • Ergonomic Workspace Adjustment: Ensure your chair, desk, and monitors are set up ergonomically to prevent strain. Invest in a quality chair and consider using a standing desk if possible.

   • Temperature and Lighting Control: Adjust the temperature to a comfortable level and use natural light as much as possible. Use blue light filters on screens to reduce eye strain, which can contribute to stress.

4. Task Transition Rituals

   • Intentional Reset: Between tasks, take 1-2 minutes to reset. Close your eyes, take deep breaths, and mentally close the previous task before starting the next. This prevents task overload and mental fatigue.

5. Mindful Communication

   • Stress Buffering Phrases: When communicating, especially in stressful situations, use phrases that buffer stress. For example, start difficult conversations with, “I’d like to discuss how we can address…” rather than jumping straight into the issue. This sets a collaborative tone.

6. Focused Breathing Technique (3-5 minutes)

   • Box Breathing: Use the Box Breathing technique (inhale for 4 seconds, hold for 4 seconds, exhale for 4 seconds, hold for 4 seconds) during moments of acute stress. This technique quickly calms the nervous system and clears the mind.

After Work Hours

1. Physical Unwinding

   • Post-Work Decompression Walk: Spend 20-30 minutes walking outside after work, ideally in a park or green space. Walking helps process the day's stress and resets your body and mind for the evening.

   • Contrast Showers: Alternate between hot and cold water in your evening shower. This method helps relax muscles, improves circulation, and washes away the day’s tension.

2. Mental Unloading

   • Journal for Reflection: Spend 10 minutes journaling about your day. Focus on what went well and what you’d like to improve. Writing down thoughts helps unload mental stress and provides clarity.

   • Wind-Down Reading: Choose non-tech, non-work-related reading material. Fiction, philosophy, or poetry can shift your mind away from work-related concerns and promote relaxation.

3. Social Decompression

   • Quality Social Time: Engage in social activities that are light and enjoyable. This could be a casual chat with a friend, playing a board game, or engaging in a hobby. Ensure these interactions are uplifting and not centered around work discussions.

   • Scheduled Disconnect: Establish a firm cut-off time for work-related communications in the evening. Let your team know when you will be offline and stick to it. This protects your personal time and mental space.

4. Sleep Rituals

   • Tech-Free Wind Down: One hour before bed, power down all screens and engage in activities that don’t involve technology. This reduces exposure to blue light, which can disrupt sleep patterns.

   • Guided Relaxation: Use a guided meditation or relaxation app for 10-15 minutes before bed. Focus on techniques that emphasize progressive muscle relaxation or deep breathing to prepare your body for sleep.

5. Next-Day Preparation

   • Pre-Bed Priority Check: Spend 5 minutes reviewing your tasks for the next day. Jot down any last-minute thoughts or to-dos to clear your mind. This helps prevent late-night rumination about work.

6. Physical Exercise Routine

   • Evening Yoga or Stretching: Engage in a 20-minute yoga or stretching session focused on relaxation and flexibility. This helps release physical tension accumulated throughout the day and prepares your body for rest.

   • Weekend Physical Activity: Incorporate a regular weekend physical activity like hiking, swimming, or cycling. Regular exercise helps maintain overall stress resilience and keeps your body in optimal condition.

Conclusion

By incorporating these specific techniques into your daily routine, you can create a balanced approach to managing stress in a high-pressure tech environment. Remember that consistency is key; the more regularly you practice these methods, the more effective they will become. Tailor these suggestions to fit your personal preferences and work style for the best results.

---

Solution Using NLP

The idea on its face is fairly simple, though I’m sure the implementation would be significantly trickier than first meets the eye. I propose that if we had an Natural Language Processing (NLP) system between the user and ChatGPT, we could generate more complex prompts to pass on to ChatGPT and therefore users would receive much higher quality responses right out of the box—without having to learn the ins and outs of prompt engineering. This would save casual users significant amounts of time and get better quality information all in one. Casual users could simply interact with ChatGPT naturally, without worrying about crafting the perfect prompt, while the system does the heavy lifting behind the scenes.

The proposed system would be simply a GUI frontend → NLP prompting engine → ChatGPT, as so:

The scope of the proposed NLP prompting engine would be as follows:

What it would do

• Parse casual users’ input to better extract intent and content

• Map unstructured phrases to structured prompt components

• Generates a detailed, prompt-engineered query based on the input

• Outputs the optimised prompt for ChatGPT to use

What it wouldn’t do

• Engage directly with users or provide responses

• Handle emotional or empathetic interactions

• Perform any tasks beyond transforming input into a better prompt

How It Would Work (brief NLP information)

There are a number of NLP semantic theories that would be used in order to transform a casual user’s prompt into an engineered one. The highlighted (different coloured) words in each sentence represent the relevant elements for each semantic point.

Intent Recognition

What’s the user’s overall goal / purpose behind their text?

Entity Recognition

Identifying the key pieces of information in a given sentence.

Sentiment Analysis

Detecting the “emotional tone” and whether the overall sentiment is positive, negative, neutral.

Contextual Understanding

This one might not be too relevant directly for the NLP engine, but it’s good to know. It’s the ability for the system to track / follow the context of a given conversation—think of it like a preference that is carried through the conversation, like not wanting to take medication, etc.

Semantic Role Labelling

Identifies who’s doing what in a sentence—as the name implies, it’s labelling the roles of different words in a sentence.

Named Entity Recognition (NER)

Nouns and more! This fella’s job is specifically to discover nouns, plus all those things that are kinda nouns but not really. Categories that fall under the job of NER include, person, location, organisation, date, time, money, percentages, events, etc.

Disambiguation

Takes things that are ambiguous and disses them until they’re the other kind of biguous. When most people hear the word, “stress”, they’ll generally think about that feeling of overwhelm—but it could also be referring to the physical kind of stress. Like the stress an elephant might put on a tub of butter by standing on it.

Coreference Resolution

Sometimes in sentences, different words are referring to the same thing. This semantic theory is what identifies these things. In the image below, “it” also refers to “stress”.

How the System Works—Casual Prompt to Engineered

The above image shows the example process from casual prompt to the engineered prompt ready to be passed on to ChatGPT.

Taking the example prompt from earlier:

Everything is too much for me at the moment, and I’m not sure what to do. I don’t want to have to take medication.

Everything is too much for me at the moment

→ Role: Psychologist who specialises in stress management

I’m not sure what to do

→ Task: Provide the user with practical, step-by-step strategies (solving their issue with perceived ambiguity on what they should do)

I don’t want to take medication

→ Exclusion: Avoid the use of pharmaceutical solutions

We would end up with the following engineered prompt:

Act like a licensed psychologist specialising in non-medication stress management. I need practical strategies for managing feelings of overwhelm without relying on medication. You will provide a list of step-by-step techniques focused on mindfulness, cognitive-behavioral approaches, and stress relief exercises. Please do not include any mention of medication or pharmaceutical solutions. Format the response in a bulleted list with short, actionable explanations.

Summary, Conclusion, and All That

This proposed NLP-engineered solution to the issue of casual prompting with ChatGPT—or most other LLMs—would significantly reduce the initial overhead for users, especially when crafting their first prompt. Certainly, there are many other aspects to take into account:

• What happens after the first prompt? How can we ensure the system continues to provide valuable assistance without requiring repeated prompt engineering?

• What about nuanced use-cases? How might the system handle highly specific or complex queries that require deeper understanding or specialised knowledge?

• What if OpenAI solves this themselves? With AI development rapidly evolving, there’s always the possibility that these issues could be addressed at the source by developers at OpenAI (or elsewhere)

• What other features should/could be introduced to supplement/compliment ChatGPT? Could integrations with other tools or additional personalisation features enhance user experience?

• What about including user preferences at startup, or as an onboarding task? Initial preference settings could tailor ChatGPT’s responses more closely to individual needs right from the start

• What about when users want ephemeral preferences? Sometimes, preferences change quickly—how might the system accommodate those in-the-moment adjustments?

• Would it be worth adding a “learning” feature? A system that adapts to user feedback over time could offer increasingly relevant responses, but at what point does this become too complex?

• When does it become more complicated than the usefulness of its features? Striking the balance between added complexity and practical value is pretty vital

• What about other prompt engineering frameworks? Exploring alternative methods might provide different approaches to improving the user experience—could be included as a ‘advanced’ option for users who are already familiar with other frameworks or just want to dabble

This was simply an idea that sprang to mind and I thought I’d write about it instead of letting it join the barren wasteland of ideas that come in and go out of my brain like a feather being eaten by the flames of a bushfire.

The Big Picture

Ultimately, it would be pretty cool for casual users to get the most out of ChatGPT without having to dive deep into the complexities of prompt engineering. This proposed system could handle all the backend niggly bits, delivering quality responses from the first interaction with minimal effort required from the user.

Your Thoughts?

What do you think about this approach to improving ChatGPT’s initial responses? Could it enhance your experience with ChatGPT? Would it make you more likely to use the tool more frequently? What other pain points do you face that I could mull over in my noggin? Whether it’s ChatGPT or anything else—let me know! And honestly, I'd welcome ideas on topics other than ChatGPT at this point—I've been a bit burned out on writing about it!

That’s it for now!

As always,

Good luck,

Stay safe and,

Be well.

See ya!

Braeden