20 Tips for Getting the Most from ChatGPT & LLM-Based Apps

ChatGPT and LLM-based apps can dramatically boost productivity, efficiency, and output. 

With the right prompts, tactics, techniques, and strategies, you can do more, work faster, and get ahead of your competitors…and your peers.

If you like the idea of automating your job with ChatGPT but aren’t sure just how to use these tools, keep reading.

This guide offers 20 essential tips to enhance your productivity and interaction with AI.

1. Understand the Context Window

ChatGPT can only remember so much.

And the longer you talk to it, the more it forgets.

Understanding the context window is crucial when working with models like ChatGPT, which have a limited “memory” for text:

• Provide context to get more relevant responses
• Break up longer texts into shorter messages
• Place instructions at the beginning or the end of a message (I saw one experiment suggesting that LLMs tended to better “remember” instructions at the end of the message)
• Periodically ask the AI what it remembers about your conversation – this will give a sense of what “context window” means

Over time, the context window and AI performance will continue to grow, so don’t get discouraged just because it can’t copyedit a 2,000-word article or write a complete research paper today…we’ll get there eventually.

2. Crafting Effective Prompts

Crafting effective prompts is the foundation of success with ChatGPT and LLMs. 

Learn to phrase prompts clearly and specifically and provide specific examples and the desired structure for your output. 

Include things like:

• The AI agent’s role
• A goal or purpose
• Your role
• What level of learner you are
• The tone/voice/reading level to write in
• Examples of the type of output you’re looking for
• How you want the output formatted

Refine your prompts and consider using custom GPTs with their instructions for repetitive tasks that need consistent output.

3. Specifying Tone of Voice

Specify the tone of voice to avoid copy that sound robotic or weird. You can even copy and paste large chunks of copy as a sample. 

With the right prompt, examples, and coaching, you can align the AI’s output with your expectations. Test different tones, and use prompt examples available online, especially those designed for copywriting and writing. 

This will help you craft responses that resonate more effectively with your intended audience.

4. Zero-shot Learning vs. Examples

Understanding the difference between zero-shot learning and providing examples is essential if you want to work with ChatGPT and LLM-based apps. 

Zero-shot learning means that the AI will respond without examples to work from – this is frequently what creates so-called “bad” output.

Providing a few examples, however, is known as few-shot learning. 

Doing this is kind of like “mini-training,” helping the AI understand your request more accurately. 

5. Formatting Instructions

Clearly outline the desired information presentation format – for instance, tell ChatGPT that you want bullet lists, markdown, plaintext, headers, subheaders, and so forth.

You can also use “variables” in your copy.

For instance, you can create a prompt that looks like this:

Extract the product information, benefits, & features from the text above. Give them to me in a bullet list that looks like this:

[Feature]. [Benefits of feature].

[Feature]. [Benefits of feature].

…

By being specific about the format, you ensure that the AI delivers the information in a way that suits your expectations the first time around. 

6. Recognizing Built-in Biases and Politics

An LLM app comes with its own set of inherent biases and political leanings programmed into it. 

Keep this in mind and adjust your prompts accordingly. 

Note that although adding specific instructions for each prompt can be tedious in ChatGPT, custom instructions and Custom GPTs can help streamline this process, since you’re only giving instructions once.

7. Using Multiple Custom GPTs

One way to automate multiple tasks is to give a lot of instructions to one agent or Custom GPT.

Another is to use multiple Custom GPTs – which is what I do.

Experiment with various GPTs to find the one that suits your needs best. 

As of this writing, ChatGPT is a frontrunner in the LLM field, so many…including me…stick with OpenAI. 

This means I can focus on my own workflows and tasks, instead of tinkering around with different GPTs – after all, they’ll all achieve parity at some point.

8. Giving Custom GPTs Access to Tools

Enhance AI capabilities by allowing access to various tools. 

Custom GPTs can use functions and tools that enable them to perform diverse tasks, such as connecting to an API to check the weather, write code, or generate images. 

The more tools an agent has, the more it can do. 

Experiment continuously to discover how much you can automate. 

The OpenAI Playground’s “assistants” and platforms like LangChain allow even greater creativity and functionality, since you can create multiple agents with a wide range of capabilities.

9. Setting Clear Goals

In your prompt, clearly define your goal and the agent’s role. 

Each agent or custom GPT needs a purpose, so make that clear by saying things like:

• “You’re a coder with 30 years of experience.”
• “Teach to a university-level learner.”
• “Your sole job is to proof for errors, correct grammatical mistakes, and fix punctuation. That’s it.”

This ensures the agent actually does what it needs to do.

10. Providing Continuous Feedback

Regularly coach the agent to refine its performance. 

Once you find the output you like, create a prompt then and there.

Or ask the AI to generate a prompt that will reproduce the output you’re getting, then put that prompt into the custom instructions.

11. Stay Updated on AI’s Developments

AI is changing rapidly and it won’t stop anytime soon. 

Though a lot of people aren’t paying close attention, that doesn’t mean it’s stopped moving. 

It’s evolving quickly, and one day we’ll wake up to a vastly different landscape. 

Since today’s early adopters will be tomorrow’s winners, it’s crucial to keep up with AI as aggressively as possible. 

Follow blogs, YouTube channels, social media accounts, and influencers – and, most importantly, use AI daily.

12. Understand Limitations

AI is not omnipotent. 

But it’s also not as useless as some people believe…

Some people outright dismiss ChatGPT as unintelligent after a few unsatisfactory responses, but others are achieving unbelievable results. 

SEO professionals, for instance, are among the most creative and innovative tech users in the world, and many of them are 10Xing their productivity with AI.

13. Consider Privacy and Stay Compliant

When working with AI, especially in a business context, understand that the data you submit might be confidential and could be used for AI training. 

Don’t submit any information you wouldn’t want companies like OpenAI or Microsoft to access. 

Being mindful of data privacy policies and the information you share is crucial for maintaining confidentiality and trust.

14. Experiment Small, then Progressively Overload

The context window for AI has a finite size, so it’s best not to overwhelm it with too much information at once. 

Start with smaller tasks and gradually work your way up to understand its capabilities and limitations better. 

Experiment with content size and use multiple messages when necessary to find the right balance for effective communication.

15. High-Level Prompt Strategies

Mastering effective communication with LLMs is key, and techniques like Chain-of-Thought prompting, Tree of Thought, and Chain-of-Verification can significantly enhance performance and reduce errors. 

• • Chain-of-Thought: 
    • think step-by-step through the process
  • Tree of Thought:

• • • Imagine three different experts are answering this question.

• • • All experts will write down 1 step of their thinking,

• • • then share it with the group.

• • • Then all experts will go on to the next step, etc.

• • • If any expert realises they’re wrong at any point then they leave.

• The question is…

• • Chain-of-Verification:

• • • [Initial prompt]

• • • List capitals of countries in Asia.

• • • [Initial response]

• • • Bangkok, Tokyo, Beijing….

• • • [Question to verify understanding/correctness]

• • • [Response]

• • • [Final answer]

Once you understand and apply these techniques, particularly in custom instructions, you can improve your output significantly…and consistently.

16. Experiment with Different Strategies

Continuously innovate with prompts, tools, and models to keep pace with the rapidly changing industry and technology. 

As the tech evolves, so should your strategies:

• Test prompts
• Experiment with your own Custom GPTs
• Experiment with those made by other people
• Try new tools and software
• Make your own tools and software

The more you learn, the better off you’ll be when the next “AI tidal wave” hits.

17. Monitor AI Evolution

Stay informed about advancements in AI and how they might affect or improve your workflow. Subscribe to:

• YouTube channels
• Email newsletters
• Blogs
• Social media accounts

Also, attend events, conferences, and seminars.

Focus on AI related to your area of expertise. 

18. Develop a Learning Mindset

AI is poised to significantly impact the world, so getting ahead now is crucial. 

Invest in learning and become an expert in AI within your industry:

• Take courses
• Get certifications
• Broaden your skill set, especially if AI is threatening your job
• Become an expert and get recognized as one

The more you can apply AI in your day-to-day job, the better positioned you’ll be for future opportunities.

19. Collaborate with Others

The AI community is overflowing with creativity and innovation. 

Engage with peers, join online communities, and follow influencers who are well-informed about AI. 

Mainstream media may not always provide the most accurate picture of AI’s potential – I often see articles on mainstream sites by people who fail to see its value. 

So stay on the right side of history: seek out those who truly understand its impact and its future. 

20. Ethical Considerations

Be mindful of the ethical implications of your AI interactions.

As AI technology becomes more pervasive, it’s crucial to consider how it’s used and the potential consequences. 

Reflect on:

• Privacy
• Data security
• Compliance
• The value you’re bringing to the world

AI is neither good nor evil. 

It is an amplifier that can be used for good or for evil, so I think it’s useful to consider how – or whether – you are adding value to the world.

• Are you developing an AI influencer that is getting kids hooked on dopamine and a brand that does nothing but drain wallets?
• Are you using deepfakes to scam people?
• Are you building an empire on teaching get-rich-quick-with-ChatGPT schemes?

Or are you building something constructive?

Conclusion

ChatGPT and other LLM-based applications are set to transform the world, one job task at a time.

We can spend forever debating the value in that transformation, but at the end of the day you’ve only got one choice: adapt or get left behind.