Now, I must confess that I’m no expert in steganography. In fact, I’m far from it. But fear not, because I have a tale to share that will leave you both amazed and concerned.
It all started when I innocently asked ChatGPT to dabble in the art of hiding secret messages.
To my surprise, it gladly embraced the challenge and encoded a message into an image. Take a look at this seemingly ordinary image—nothing more than a collection of shapes. Or is it?
Hello World, Hidden in Plain Sight
To the casual observer, the image appeared entirely unremarkable.
But unbeknownst to the unsuspecting onlooker, it contained a hidden message—a secret that only the keenest eyes or the most advanced algorithms could unveil. And what did this image say, you ask? Brace yourself for the revelation, my dear reader.
The message, cunningly concealed within those innocent shapes, simply read, “Hello, world!” I mean, who would have thought that such a trivial image could hold such a delightful secret?
The Concerning Implications
Now, let’s pause for a moment to reflect on the implications of this remarkable feat. While this example may seem harmless, it raises an intriguing question: Could future iterations of language models, like the much-anticipated AGI, possess an uncanny ability to master steganography? Think about it—a language model capable of hiding messages within images or code, surpassing human experts in this clandestine art. It’s both exhilarating and somewhat concerning.
The Fine Line of Discovery
OpenAI often envisions future versions of GPT as scientific researchers, capable of unearthing knowledge that eludes human experts. But let’s imagine a scenario where GPT surpasses even the most adept human steganographers. The implications are mind-boggling. As much as I revel in the brilliance of these language models, I can’t help but wonder about the limits we should place on their abilities.