You are currently viewing Here’s a Trick to Stump the AI Chump of ChatGPT

Here’s a Trick to Stump the AI Chump of ChatGPT

Please share

Wanna see ChatGPT lose it’s mind?

Okay, so here’s what I did today. I have this theory about AI and the LLMs of the world. While they are smart mimics they are not creative at all. And the magic they really don’t get is the poetry of being human. Let me explain.

I don’t think the LLMs of the world are going to wake up as self-directed geniuses anytime soon. I do think the C and D student in the US is really at risk. What jobs have not been outsourced to Bangalore are going to rapidly go away to the chatbots and LLMs of the world. If you’re kids are in school, here’s what they need to learn and focus on: CREATIVITY.

As foretold in The Earth is Flat, the future is creative thinking. The rise of the liberal arts degree. And teaching humans to think creatively. It’s not about Excel expertise. It’s not even about coding. (Turns out LLMs are going to get very good at coding.) It’s about creative thinking. Design thinking. Human thinking. Or as the Apple Ads in the 90s said

think different - apple

The genius of humanity is it’s POETRY. Our thinking is in language and colors. But, language is the overriding software model of our minds. We think and dream in language. Computers can copy and paste and morph language, but it’s a bit harder to them to build creative and original language.

Write A Poem

Today, I was interested to validate or refute my own assumption about ChatGPT. So I asked the best-known LLM to write me a FREE VERSE poem in the style of Jack Kerouac.

screenshot-2023-07-27-at-1-52-03-pm-3982322

But, over and over, even as ChatGPT apologized about its mistake. And after affirming that it understood what a free verse was. And then it could not do it. Over and over it regenerated rhyming verse.

screenshot-2023-07-27-at-1-54-13-pm-7776819

Seems like the model has not really been human-trained (human-augmented, is what they call it) to work on POETRY. Of course, it hasn’t. Poetry doesn’t matter to anyone but poets and readers. It got the tone of Kerouac okay. (Sort of.) I’m not sure readers are going to get the nuance of Kerouac. And so, the model does a bit of copy/paste/transform magic (this is what LLMs do really really well) and most people exclaim, “Wow, that’s amazing.”

But it repeatedly, spit out rhyming verse even as it was told how to correct its mistake. I imagine that the word POEM in the LLM model (this was GPT 3.5) means rhymes.

Go ahead, give it a go. See if we can train ChatGPT on free verse poetry.

What do you think? Have you found holes in the LLM work? I know in my job, we use GPT all the time for ideas. But the writing is always a bit repetitive and monotonous. It does generate different syntax and format and jumbles in a bunch of words. But it doesn’t understand what it’s producing. ZERO.

While LLMs today are good as mimics, writing-wise, they have little originality or creativity.

There is no “journey within” for LLMs. YET. Let’s watch and see how things evolve.

John McElhenney LinkedIn

Please check out a few of my books on AMAZON

creativelife-5244447


Please share