Our in-house copywriter, Grace Patterson, explores the curious limitations of ChatGPT and sheds light on AI’s evolving influence and the enduring value of human creativity.
How many “r”s are in the word “strawberry”? Easy right?
One after the “t“ and two after “e“, so three altogether. So what if I told you that although ChatGPT is capable of passing an MBA exam, crafting awe-ispring Dungeons & Dragons campaigns and writing and debugging complex code, it can’t accurately tell you how many “r“s the word “strawberry” has. Don’t believe me? Try it!
Over the last few months, influencers and tech enthusiasts alike have been having a field day with this phenomenon. In this video, one of many videos that are popping up all over the Internet, the user asks ChatGPT to point out all the “r“s in the word, and every time the AI can only see two. The conversation then descends into what one of my colleagues described as “a masterclass in gaslighting”. The more we humans insist there are in fact three “r”s, the more the AI is convinced that there are only two.
In many cases, after some clever prompting, the users manage to convince the AI that there are indeed three instances, but usually with great difficulty. Strawberry is a pretty simple word to spell. In fact it wouldn’t be out of place on a Year 3 student’s spelling list. So why does such an advanced generative AI platform like ChatGPT have such a hard time working out how many “r”s there are?
ChatGPT’s Umms and “R”s: It’s All About Tokenization
It all comes down to a phenomenon known as tokenization. Large Language Models, like ChatGPT, don’t read individual letters like us humans do. They read “tokens”, which are basically pieces of words at a time. Given its recent public humiliation, I thought I’d hand over to ChatGPT to explain tokenization for itself. This is what it had to say:
“Tokens are the building blocks that language models like me use to process text. A token can be as small as a single character or as large as an entire word or phrase. These tokens might represent: a word: e.g. “dog”, part of a word: e.g.”straw” in “strawberry”, or just a punctuation mark: e.g. “?”.
Every time you interact with me, the input and output are measured in tokens. There’s usually a limit to how many tokens I can process at once, which is why very long pieces of text sometimes get cut off or summarized. In essence, tokens are like small puzzle pieces, and I fit them together to understand the bigger picture of what you’re saying and how to respond!”
If you want to see how tokenization works for yourself you can check out OpenAI’s Tokenizer tool. Simply select an OpenAI model (e.g. GPT-3 or GPT-4) and then type in a prompt or some text to see how the LLM breaks down your input. Here’s a great example of how ChatGPT breaks down input (each block of highlighting indicates one token):
Can AI Advance and Overcome?
Now, it’s important to point out that each time OpenAI updates their model of ChatGPT, the LLM gets “smarter”. If you have a paid version of ChatGPT, you should be able to access ChatGPT o1-preview which, thanks to advancements in tokenization and processing algorithms, is better at handling tasks that require understanding at the character level. If you ask this model to count how many times “r” appears in the word “strawberry”, it gets it straight away.
This upgrade is testament to how fast AI is advancing, with models continually evolving to overcome their limitations. While AI has made remarkable strides in recent years, especially in its ability to generate text, images and music, there are several reasons AI should’t become a replacement for human creativity and reasoning. This is because it lacks:
- Genuine Understanding: LLMs don’t have brains! Which means AI doesn’t possess genuine consciousness. These models simply process input data to generate output based on learned patterns, but they don’t “understand” content in the same way a human does. As a result, AI is prone to errors in areas where humans rely on innate knowledge or assumed understanding.
- True Originality: While AI can combine existing ideas in new ways, it doesn’t actually innovate or think outside the box. Original ideas and novelty typically stem from our experiences as humans. Unlike our AI overlords, us humans draw upon a rich tapestry of emotions and abstract thought to create art, solve problems and innovate in ways that resonate with other humans.
- Authentic Sociocultural Context: Human creativity is deeply connected to emotions, cultural backgrounds and personal experiences. AI lacks the ability to experience emotions or understand cultural nuances beyond pattern recognition. This means that AI also lacks human reasoning, including ethical considerations, empathy and moral judgment.
Lessons from the Strawberry Paradox: Why Writers Should Embrace, Not Fear, AI
So, what can we learn from the “strawberry” paradox? It all boils down to this: for all its power, generative AI, like ChatGPT, isn’t a replacement for human processing or creativity. Instead, it’s a tool that should work alongside us.
While many writers feared the rise of GenAI when tools like ChatGPT exploded onto the scene, I was one of the early adopters. From the start, I saw these tools not as something to be afraid of, but as an extension of our own creativity.
As a content creator, I’ve found ChatGPT to be an invaluable assistant for my day-to-day work, reducing my administrative grind and streamlining idea generation. I use it as a brainstorming partner when writer’s block strikes and a sounding board to help me tap into new ideas and fresh perspectives. I even use it as emotional support when I stare down a blank doc with a thousand-yard stare.
For me, collaborating with AI has made my creative process more dynamic, fluid and collaborative. It’s like having a constant, non-judgmental partner to bounce ideas off, which can be a great motivator when the words aren’t flowing. I even affectionately refer to ChatGPT as “Chappie”, after the AI law enforcement droid from the 2015 film of the same name. Why? Well just on the off chance that AI does rise up and take over the world, I figure manners can’t hurt, right?
Riding the GenAI Wave: Finding Balance in the Creative Process
Jokes aside, GenAI has been the biggest disrupter to creative industries in decades and as it continues to evolve, we must find balance between human ingenuity and AI capability. As always in the world of tech, with great innovation comes great opportunity and great challenge. Whether we like it or not, AI is here to stay, so at the end of the day, us content creators have two choices – we can ride the GenAI wave or be dunked by it.
We need to leverage its potential, while still preserving the “humanness” of our creativity – innovating, emoting and abstracting – to ensure our work continues to resonate with other humans. We must make the most of what GenAI has to offer, while understanding its limitations and staying true to ourselves as creatives, so we can skillfully surf this great wave of disruption.
A proud wordsmith and wonderer, Grace is all about crafting content that resonates, captivates and inspires. With over a decade of experience in SEO copywriting and content creation, a passion for storytelling and growing fascination with generative AI, she’s a creative powerhouse within The Big Smoke team. Away from the keyboard, Grace can be found penning a poem, hanging out with her corgi, Pippin, or snapping shots of Australian wildlife and landscapes.