“Won’t someone rid me of this meddlesome AI?”
It’s the million dollar question we could hear website owners far and wide asking themselves as they pulled their hair out. Chat GPT and other large language models (LLMs) like it are producing more content than we could ever imagine. And with another means of production comes another fun way of spamming us all.
A Successful Failure
Let’s first skip to the end. We will not be implementing any kind of anti-LLM feature into Stop Spammers or Stop Spammers Premium. We were going to hook into an AI detection tool and then offer a variety of features to support your anti-spam needs. The very nature of detection is where we were hung up.
ChatGPT’s introduction fulfilled a huge need, but it’s wide adoption created another. The development would mark another chapter in the wild west of content creation, and we desperately wanted to decipher human from artifice. In the void, several companies stepped up, including OpenAI–the makers of ChatGPT. They claimed they would score content as being within a range of definitely human to definitely AI.
It was an alarming time in the frenzy that followed this new Gutenberg on steroids. And the adoption of AI detection models was every bit as quick. Professors started to leverage these dubious models to determine what students to fail.
It turns out, in our haste to find a way to detect AI, we turned to a solution that doesn’t work. Evidence is beginning to show that AI detection tools don’t stand much better chance at detecting AI from humans than simply flipping a coin. The tools misfire especially dramatically when reviewing content written by someone in a language that’s not their native tongue. Seeing the writing on the wall, OpenAI shut down their own detection model with little fanfare.
It gives me no pleasure to say this, but these “tools” are little better than snake oil. Like most of us, I wanted them to be the silver bullet. I wanted Stop Spammers to have an extra layer of security that tackles artificial intelligence. But we must abandon any hope of a detection tool for now. With the introduction of GPT4, these models are going to on the whole, trend to more sophisticated content creation.
Pursuing a Different Direction
We still could have implemented a detection tool hook. AI is en vogue, and saying that we addressed LLMs would surely generate clicks. But we know the engine behind the feature(s) is fatally flawed, and it would be deceptive to deploy something purely for good buzz. Instead, we are circling back to bread and butter measures. Many are low hanging fruit. In the coming months, you’ll see a whirlwind of new features, several of them will be in the free version of stop spammers.
Spammers and scammers have calling cards because they have certain aims, and because of that, we are able to continue refining our checks. By the end of the year, Stop Spammers will be even more robust than ever before.