As an entrepreneur and consultant, I’ve spent years designing systems and implementing machine learning (ML) strategies using pre-trained models like those provided by OpenAI. During this time, I’ve talked to many business leaders seeking to improve their businesses with artificial intelligence (AI) and have encountered a few common misconceptions about what Large Language Models (LLMs), like ChatGPT, can and can’t do.
Since the release of ChatGPT, I’ve been asked even more frequently about the capabilities of these tools. These models are awe-inspiring, but there are also many limitations. However, with the right lens, what many people consider limitations become features of this once-in-a-generation technological development.
In this blog post, I’ll share some of the most common misconceptions about ChatGPT and LLMs and provide insights on how you can best utilize them in those same contexts. So, let’s dive in and separate fact from fiction regarding ChatGPT and LLMs!
The reality is that ChatGPT’s writing often feels overly generic and can’t replace the creativity and nuance of human writing. Furthermore, using ChatGPT to write high-stakes content is a risky move, as it’s likely that someone will discover ChatGPT’s authorship. OpenAI is actively working on detecting this type of content, so it’s best to avoid relying on ChatGPT for writing anything substantial.
Search Engine Optimization (SEO) is a frequent use case where people seek to generate content using ChatGPT. While many SEO experts worry about the effect easy access to new content will have on the space, it’s only a matter of time before the low-quality content begins to get filtered out. So while you can create new blog posts solely with ChatGPT, long-term negative associations with low-quality content will likely outweigh the short-term benefits of producing low-quality content.
Instead, it would be best to use ChatGPT as an aid, much like using sources from a search engine. When the stakes are low and time is of the essence, ChatGPT can be a helpful tool for generating quick and straightforward responses. However, relying on human writing style and expertise for writing important content remains the best option. I recently wrote more in-depth about why I wouldn’t use ChatGPT to write a 2,000-word article.
How to use ChatGPT in your writing:
✅ Drafting responses to emails
✅ Outlining ideas for your content
✅ Writing a summary for internal use
How not to use ChatGPT in your writing:
❌ Writing a blog post for SEO
❌ Writing a research paper
While it’s true that ChatGPT can assist with simple coding tasks, there is a false belief that it enables anyone to write code. In reality, it takes a deep understanding of programming to build a standalone application, even with the help of ChatGPT.
The value of using ChatGPT for coding decreases as the complexity and nuance of the task increase. So I don’t recommend using it for complex and nuanced tasks. Instead, ChatGPT is best for coding that would otherwise be copied & pasted, like generating “boilerplate” code for a new application or creating a function for a well-known algorithm.
While generating code from scratch may sound like the end-all-be-all of AI features for coding, there are other ways that AI can help in the process. In my personal experience writing code, and the general sentiment of professional programmers, GitHub Copilot is the most highly regarded of the AI tools for coding for the ability to assist with writing code, albeit one line or block of code at a time. At the time of this writing, given its utility, failing to use Copilot is as much of an oversight as failing to use Grammarly in your writing.
While it’s impressive that ChatGPT can write code at your command, given contextually aware tools like Copilot, it’s unlikely that ChatGPT will become the most impactful AI technology in the programming field.
Logic & Reasoning
While it may seem like the bot is using logic or reasoning to respond to user inputs, this is a false belief. In truth, ChatGPT predicts the most likely word to come next based on its training data and user inputs.
It’s important to understand that ChatGPT’s responses are always predictions, even if they may sound confident and thoughtful. As a result, its accuracy has limits when responding to factual requests. However, despite this limitation, ChatGPT is still a revolutionary technology that has the potential to change the way we do things as long as we utilize it correctly.
The truth is that this lack of grounding in reality, known as hallucinations, may be one of the most remarkable features of ChatGPT and language models in general. The predictions it makes are the basis of imagination and creativity. If it only responded with excerpts from articles, then it would just be another search engine and would lack the ability to help you explore your imagination.
So while ChatGPT may not be the best research assistant for a thesis paper, its positive impact may be even more significant than such a tool, given its ability to be your brainstorming partner, thus enhancing the creativity and imagination that make humans unique.
Perhaps an extension of its limits in logic and reasoning, another common misconception is that ChatGPT can solve complex math problems. However, while ChatGPT may appear capable of solving simple arithmetic and some algebraic equations, it regularly fails with more complex math problems.
This inability to solve math problems may seem paradoxical because our calculators and computers have been successfully helping us solve math problems for over 50 years. So why can’t AI, running on a computer, solve the same equations?
This paradox is an excellent opportunity to recognize the fundamental shift in how the technology underlying our interactions with computers has changed. For example, it’s common for humans to refer to others based on their abilities, like being a “math person” or a “creative.” Each designation comes loaded with expectations we have of those types of people.
Until now, computers have always been the “math person” who could be relied on for calculations, but not the “creative” you would go to for works of art. Technology like ChatGPT is our first glimpse at what it might look like if computers played the role of “creative.”
To understand ChatGPT’s math capabilities, imagine you need a math tutor. But for the sake of this discussion, you decide to go to the “creative” person for help. Assuming they know the math required, the “creative” will have particular strengths in helping you, like communicating the algorithm in easy-to-understand language instead of mathematical notation. But coming up with an exact answer is still something they rely on a calculator to complete.
Suppose you need to use ChatGPT for math assistance. In that case, ask for instructions on solving a math question rather than relying solely on its final answer. For example, I have witnessed ChatGPT providing the correct formula while giving a wrong answer after the equals sign. So, it’s best to use ChatGPT for math assistance with some caution. And always use a traditional calculator to check your final answer!
Fine-tuning & Specific Use cases
It’s important to understand that ChatGPT is not a standalone solution for all AI needs. While it’s a highly competent tool, you can combine its technology with other AI models to achieve more impressive outcomes with less complexity and cost than fine-tuning.
Many businesses want to refine their abilities to provide more context-specific responses, like answering based on corporate data and documents. It’s a common belief that fine-tuning ChatGPT is the best way to achieve this. But there are alternatives to fine-tuning that enable you to achieve the same, and likely better, results than you would get with a single fine-tuned model. These alternatives are even more important early in the product development cycle, where flexibility and speed are keys to success.
It’s also important to understand that ChatGPT currently cannot utilize fine-tuning or be used as an API to build products. Instead, it’s a product created by OpenAI to demonstrate the capabilities of their language models. So if you’re looking to develop a similar effect, you will have to use the other available OpenAI API products. However, recently due to demand, OpenAI has indicated that it will soon release an API product based on the popular ChatGPT. But you should note that some of the other most impressive API models, like
text-davinci-003, cannot be fine-tuned. So if you want the high-quality conversational experience provided by these models while still being able to answer accurately based on the context of your corporate documents, then understanding how to integrate additional AI models, like embeddings, into your solution is essential to maximizing your success.
ChatGPT is not a standalone solution for all AI needs. Instead, it’s one tool in the AI toolbox. While LLMs may not be the best research assistant for a thesis paper or the go-to solution for complex coding tasks, they continue to revolutionize how we do things. Their tendency to hallucinate may even be recognized as one of their most remarkable features, as it can spark imagination and creativity on a massive scale.
By dispelling some common misconceptions about ChatGPT, I hope you feel empowered with the insights necessary to utilize this technology to increase productivity and improve your business.