Everything I’ve Learned About AI

In March 2023, an exciting transformation occurred within my role at Carrot. We made a remarkable addition to our team by bringing on board an exceptional COO with the expertise to enhance our business’s performance. Consequently, my position underwent a significant shift, propelling me into the role of VP of Strategy.

This pivotal juncture aligned perfectly with a flurry of captivating AI news releases, capturing my undivided attention. The world of artificial intelligence beckoned me, and I eagerly delved into its intricacies. In this post, I aim to consolidate the wealth of knowledge I have acquired, highlighting key aspects of AI that have sparked my enthusiasm and addressing certain concerns.

It is important to acknowledge that I am not an expert just yet. Still, my passion for AI has driven me to invest more than 50 hours immersing myself in research, devouring informative articles, engaging with insightful YouTube videos, and experimenting with over 50 AI tools. Therefore, I am more than a casual observer of this remarkable technology.

Now, let’s dive into the current AI landscape and explore the trends that have caught my attention.

1. Reinforcement Learning: One notable trend that has been making waves is advancing reinforcement learning techniques. Through this approach, AI systems can learn and improve by interacting with their environment, mimicking how humans acquire knowledge through trial and error. The potential applications of reinforcement learning are extensive, ranging from optimizing complex processes to enabling breakthroughs in robotics and autonomous systems.
2. Natural Language Processing: The natural language processing (NLP) field continues to evolve astonishingly. AI models are becoming increasingly proficient in understanding and generating human language, significantly improving machine translation, sentiment analysis, chatbots, and virtual assistants. The ability to comprehend and communicate in a natural language is a game-changer, paving the way for more intuitive human-AI interactions.
3. Ethical AI: As AI becomes more prevalent, ethical considerations are gaining prominence. Addressing bias in AI algorithms, data privacy, and transparency has become crucial. A pressing need is to develop robust frameworks and guidelines to ensure responsible AI deployment. It is heartening to witness the growing emphasis on ethical practices and the concerted efforts of researchers and policymakers to navigate the ethical challenges of AI.

AI is NOT like crypto.

I’ve heard some people dismiss AI’s impact on our society. “Oh, AI, that is just this year’s cryptocurrency fad.” This couldn’t be further from the truth. Unlike blockchain, NFTs, crypto, and other recent tech advancements, AI has utility and is WAY easier to use. AI saves time and works after 1-2 clicks.

A better analogy would be when the internet launched or Google Search came online. I am 100% convinced that AI isn’t going anywhere and will transform our lives (hopefully for the better), and AI is imperative for employees to learn how to use it, or they risk being replaced by it.

ChatGPT & Bard

Most of my AI experience is with text-based tools like Open AI’s ChatGPT and Google’s Bard.

ChatGPT is an extensive language model (LLM). Massive volumes of data are used to train Large Language Modeling (LLMs) to anticipate the next word in a phrase correctly.

It was revealed that increasing the quantity of data enhanced the language models’ performance. Stanford University states:

“In order to train GPT-3, 570 terabytes of information and 175 billion parameters were utilized. In comparison, its predecessor, GPT-2, has only 1.5 billion parameters, which is more than 100 times less than its successor. This large change in size has a profound effect on the behavior of the model, enabling GPT-3 to complete tasks on which it was not expressly instructed, such as translating words from English to French, with extremely little or no training samples at all.”

Other AI generative models exist for things like video, images, music, and more. But for now, most of this blog post will cover text-based models. This brings me to my most significant insight…prompts.

It’s All About The Prompts

Like many new ChatGPT users, I would type in “Write me a blog post about home prices in America.” The model would give back a vanilla, formulaic blog post that was a good starting point but not worth posting online.

I started hearing about prompt engineers and people working hard to “uncover” what prompts specific tools or people were using to get results back from AI.

Prompts are where the magic happens. The better you prompt the AI, the better your results will be. So far, it’s the highest leverage value I’ve found in my AI deep dive.

12 Types Of Useful Prompts

1. Explanation prompt: These types of prompts require ChatGPT to clarify a concept or process. For example, “Can you describe the [technical term] in simple terms?” or “How does [product/service] function?”*
2. Fill-in-the-gaps prompt: These prompts enable you to enter certain information into a statement or question. For example, “My company provides [product/service] to help customers [achieve goal/benefit].”
3. Yes/No prompt: These prompts require ChatGPT to respond with a simple “Yes” or “No.” For example, “Do you believe [industry/niche] is becoming more popular?” or “Have you used [product/service] in the past?”
4. Multiple selection prompt: These offer a set of alternatives for ChatGPT to select from. For instance, “Which [product/service] do you prefer?” or “Do you like [Option A] or [Option B]?”
5. Ranking prompt: These prompts require ChatGPT to rank a series of things in order of importance or preference. For instance, “Rank these [products/services] from most to least popular” or “Which [product/service] do you think provides the most value to customers?”
6. Unrestricted prompts: These types of prompts allow ChatGPT to produce a wide array of responses. For instance, “Share your experience with [product/service]” or “What are your views on [industry/niche]?”
7. Instructional prompt: These prompts require ChatGPT to provide instructions or guidance on a specific task or process. For example, “How do I [perform a specific task]?” or “What are the steps to [complete a specific process]?”
8. Situation-based prompt: These types of prompts require ChatGPT to provide a response based on a specific situation or circumstance. For example, “What would you do if a customer complained about [issue]?” or “How would you manage a situation where [problem] emerged?”
9. Comparative prompt: These prompts require ChatGPT to compare or contrast two or more items. For instance, “What are the distinctions between [product A] and [product B]?” or “Which [marketing strategy] do you believe is more effective?”
10. Forecasting prompt: These prompts require ChatGPT to make a prediction or projection about a future event or trend. For instance, “What do you predict will be the most significant [trend/issue] in [industry/niche] over the next five years?” or “How do you anticipate [product/service] will perform in the market next year?”
11. Opinion prompt: These prompts require ChatGPT to provide an opinion or perspective on a topic. For instance, “What is your take on [controversial topic]?” or “Do you think [new trend/technology] will be successful in the market?”
12. Feedback prompt: ChatGPT must provide feedback or suggestions on a product, service, or idea. For example, “What feedback do you have for my [website/marketing campaign]?” or “Do you have any recommendations for improving [product/service]?”

Here is an example of a ranking prompt being used.

Helpful Tips & Tricks To Get Better Results

If you want to use AI like a pro, here are some common ways to leverage ChatGPT and Bard better.

01. “Act as” sentence structure.

Crafting an engaging and immersive conversation with the AI language model, ChatGPT, relies heavily on formulating effective prompts. One valuable technique to achieve this is the “act as” hack, wherein ChatGPT is directed to assume a specific role or persona within the dialogue. This approach proves particularly beneficial for simulating real-world scenarios and fostering a captivating experience. For example, one may instruct ChatGPT to embody the role of a detective and unravel a fictional crime, or act as a travel agent and provide personalized vacation recommendations based on user preferences. By utilizing the “act as” hack and providing a detailed description of the desired role or persona, users can tailor ChatGPT prompts to suit their specific requirements. This technique offers limitless possibilities and is an indispensable tool for professionals seeking to create immersive conversations with ChatGPT.

Act as a Tech Reviewer:

I want you to act as a tech reviewer. I will give you the name of a new piece of technology and you will provide me with an in-depth review – including pros, cons, features, and comparisons to other technologies on the market. My first suggestion request is “I am reviewing iPhone 11 Pro Max”.

Act as a Social Media Manager

I want you to act as a social media manager. You will be responsible for developing and executing campaigns across all relevant platforms, engaging with the audience by responding to questions and comments, monitoring conversations through community management tools, using analytics to measure success, creating engaging content, and updating regularly. My first suggestion request is “I need help managing the presence of an organization on Twitter to increase brand awareness.

02. Make ChatGPT Sound Like You

Often a ChatGPT response sounds too formal, formulaic, or obvious. You can change this by uploading a couple of paragraphs of your writing that capture your tone, style, and personality. Then ask ChatGPT to write something as the author above would.

Prompt: [Insert Text with your Writing Style]

Analyze the writing style and write about [Topic] as the above author would write.

03. Forms of answers hack

Paragraph or essay-style answers:

This form is ideal for providing in-depth explanations or analyses of a topic related to ChatGPT or freelancing. It allows for a more nuanced exploration of a concept or idea and can be particularly useful for complex or abstract subjects. For example, a chapter on the ethics of using ChatGPT in freelancing could benefit from a paragraph-style answer that delves into the potential moral implications of relying on AI technology.

Bulleted lists or numbered lists:

Lists are useful for providing concise and straightforward information about a particular topic. They can break up large text sections and make information easier to read and understand. For example, a list of the top five ways freelancers can use ChatGPT to save time and increase efficiency could provide valuable insights for readers.

Tables and charts:

Tables and charts are useful for presenting complex data or information clearly and organized. They can be handy for comparisons, statistics, or other numerical information. For example, a chart comparing the cost savings of using ChatGPT versus hiring a human assistant could help readers understand the technology’s financial benefits.

Step-by-step instructions or tutorials:

Step-by-step instructions or tutorials are an excellent way to guide readers through a specific process or task. They can help teach technical skills or other practical applications of ChatGPT. For example, a tutorial on integrating ChatGPT with popular content management systems like WordPress or Squarespace could be helpful for readers.

Code:

Code is a form of explanation that is ideal for technical or programming-related topics. It involves providing code snippets or examples to demonstrate how a particular process or task can be accomplished using ChatGPT. For example, code examples on how to integrate ChatGPT API with Python or Java programming languages could be helpful for readers who are interested in developing their ChatGPT-powered applications.

04. Training The AI (not as scary as it sounds)

While ChatGPT is an advanced AI model trained on vast amounts of data, fine-tuning its responses to more specific prompts can be fine-tuned by providing additional examples or data. This process is known as mini-uptrain- ing, and it can be a powerful tool for improving the accuracy and relevance of ChatGPT’s responses.

By providing relevant examples or data, you can help ChatGPT understand the specific nuances of a particular topic or industry, allowing it to provide more tailored responses. For example, a marketing agency might provide ChatGPT with examples of successful ad campaigns or branding strategies to help it better understand the needs and goals of its clients. This can ultimately lead to more effective and targeted marketing campaigns.

Simply provide additional examples or data to ChatGPT when designing your prompts. This can be done by including specific examples in the prompt text or by giving examples. Best to add it in [X] or <X> form. You can unlock the full potential of ChatGPT, creating more accurate and relevant responses.

Example of such prompt:

• Provide me a list of great keywords for ChatGPT SaaS product X. Use the keywords below as a good example: [List of keywords]
• I’m writing an article, “How to build a Chrome extension.” Write me 15 unique and authentic tips on building, testing, and marketing a Chrome extension. Use the list of tips below as an example and design 15 new ones: [List]

Wrapping It Up

I could easily write 10,000 more words on what I am learning with AI. For now, I’ll leave these findings here. I am learning more and more each week and testing these prompts with skills like content marketing, ranking websites, and converting traffic.

The output of this labor is going into improving copypress.ai and the content and product at Carrot.com