AI or Not AI? That is the Wrong Question
AI-generated image, created on Recraft.ai

AI or Not AI? That is the Wrong Question

I read a reddit post by a fellow freelancer yesterday that irked me.

The freelancer's client ran their work through an AI detector, which flagged a sliver of the piece as potentially being generated by AI. The freelancer is adamantly against AI — they won’t even use it for ideation, research, outlines, or anything else — so the piece was 100% human.?

That's also supported by the fact that a mere 4% of their work tested positive for AI. Yet, the client made them revise it until the detector, ZeroGPT, output a clean 0%.

Utter nonsense.?

Let’s discuss why. But first, some context: I use ChatGPT. I think it’s a practical tool.?

The operative word in that sentence is tool. It’s not half bad at creating outlines and clever headlines (somewhat hit or miss), compiling audience personas, and serving generally as a sounding board for ideas. But it’s not a replacement.?

You can tell the difference between AI-generated text and good writing, because one is fluff and one is good writing.??

So, I decided to run a little experiment, running a variety of my old work (pre-ChatGPT) through ZeroGPT to see what the so-called AI “detector” could detect.?

The post: How Long It Could Take Your Portfolio to Recover From the Next Bear Market (published 1/19/2022)

The result:

And you’re likely (definitely) wrong.?

Dissecting ZeroGPT’s analysis, it doesn’t appear to be a fan of whenever I used numbers, which this article was full of.??

Even so, I had plenty of figure-less text that triggered the system — like my entire conclusion.?

Let’s try another.?

The post: 5 crypto scams to know before you start trading coins (published 9/16/2022)

The result:

And, again, wrong.?

This time, ZeropGPT flagged the whole introduction.?

I ran several old client pieces that also predated ChatGPT through the detector too, and the results were the same: anywhere from 20% to 60% of the copy was flagged. (Considering it was ghostwritten, I won’t share the actual content, but rest assured it was 100% written by yours truly.)

What can we take away from this charade??

  1. Just because AI detectors flag copy doesn’t mean it’s AI-generated.
  2. Just because AI detectors flag copy doesn’t mean it’s bad.
  3. Just because AI detectors flag copy doesn’t mean you should revise it until it’s “human.”??
  4. AI detectors suck.?

I like another redditor’s response to the situation: revising toward some arbitrary goal is a worthless, out-of-scope exercise. It’s a fruitless game with inconsistent and inaccurate rules. And if a client is too thick in the head to realize that, then perhaps it’s time to move on from said client.?

Maryna Parasiuk

Посада Outsource Manager у Boosta

9 个月

ZeroGpt is a crazy checker. You may add extra letter and receive completely different score.

回复
Thilini Wijesinhe

Content for financial businesses | Building a remote business

9 个月

I haven't had any clients say this. But I've tested AI detectors with my own writing (and certain parts were flagged as AI generated).

回复
Erika Domingo

I design systems that let entrepreneurs clock out | Workflows Strategist | Dubsado Specialist | Notion Consultant

9 个月

I feel like the vitriol against AI is so misplaced because the conversations around it lack the nuance it demands. There's this misconception that writers just prompt something into the LLM and it comes out with a perfect output that the reader can't detect when as you've mentioned - it truly is just a tool. The ideas, the expertise, and the core of your piece come from you - not the LLM.

回复
John Nusen

Freelance Email Marketing Specialist | Grow your Business with Email Marketing

9 个月

I am going with #4 AI detectors suck You could add #5 Humans who use AI detectors suck more. It reminds me of a certain test that most of the world took in 2020 that turned out to be flawed, which caused the same majority of folks to take yet another flawed test that turned out to be lethal in the end. The way of the dodo bird is upon us, that is all.

回复
Farnaz Kia

Building Partnerships That Drive Growth | SEO Content Nerd | Crafting Content That Ranks | BrandWell AI, WorkMind AI, Chain AI

9 个月

Here is a fact about AI detectors: AI detectors can tell whether content sounds robotic, but they can't say if it is AI-written. This can lead to false positives, especially for non-native English speakers, whose writing may seem more robotic than that of a native speaker.

回复

要查看或添加评论,请登录

Carter Kilmann的更多文章

社区洞察

其他会员也浏览了