Next-Level ChatGPT: Can It Read Your Mind?

Photo by Anja Bauermann at

As I write this column on a bitterly cold December day, the second day I’m without water in my home because the pipes are frozen, it was only natural to ask ChatGPT ( to write a poem about why I’m tired of this wintry weather.

In seconds it came up with this:

The icy winds are blowing strong,
The sky is gray and dull all day long.
I bundle up in coats and scarves,
But still I shiver and feel the cold’s sharp barbs.

I trudge through slush and frozen snow,
Hoping for a warm place to go.
But even inside, the air is nippy and chilled,
Leaving me feeling empty and unfulfilled.

I long for sunshine on my face,
For the warmth of summertime’s embrace.
I’m tired of the cold, it drains my soul,
Leaving me feeling blue and out of control.

So here’s a little prayer to the weather gods above,
Please send some warmth & sunshine with your love.
I’m tired of this cold, it’s time for a change,
Bring on the heat, let summertime reign.

ChatGPT nailed it. It’s a delightful poem, and my online check for plagiarism  tells me it’s original.

I first wrote about GPT-3 a couple years ago, and late last year GPT-3.5 was released, taking artificial intelligence language generation to a new level. ChatGPT is a public demo of this new version, with GPT-4 expected soon.

As we’ve discussed in previous columns, these AI text generators are fed gobs of text and use “machine learning” to assimilate the text. What do they learn? They simply learn what words typically follow the previous words. It’s similar to the predictive text entry on a smartphone in which it anticipates the next word—but is hugely more powerful.

ChatGPT wasn’t taught rules of rhyme and meter. Rather, this new version was fed poetry and used neural network technology to learn the principles, such that it can produce poems on demand.

As explained to me by ChatGPT, “As a large language model, I have been trained on a wide variety of text data and can generate natural language text in a variety of styles and formats. This includes the ability to generate poetry, as well as other types of text such as stories, articles, and descriptions. . . . While I can generate poetry and other types of text, it is important to note that I do not have personal experiences, emotions, or creative intentions like a human writer. My responses are based on the patterns and structures I have learned from the data I was trained on, and I do not have the ability to add personal perspective or intention to my responses.”

After its release on November 30, ChatGPT was so impressive that it generated huge publicity. People were astonished by what it could do. Wary of the possibilities, teachers tried entering the essay assignments and exam questions that they gave their students—and ChatGPT earned good grades. Uh oh.

OpenAI, which developed ChatGPT, has actually created a tool that can check whether text has been created by AI or a human ( It correctly identified the quotation above as created by AI, but when I gave it the first two stanzas of its poem, it said it was “real.” Uh oh.

The range of things ChatGPT can do is astonishing, from writing jokes to explaining complex topics. Perhaps most impressive is its ability to write computer programs. For example, it wrote code for a website in response to this query: “Write code for an ecommerce website using html and CSS. Include navigation bar, body section, and products with prices.”

Still, ChatGPT has limitations, as I found. Most crucial is that it has no ability to determine the correctness of what it writes. When I asked for a history of Fairfield, it wrote that the town was founded by abolitionists. But local history teacher Lawrence Eyre told me in an email that that wasn’t quite accurate. He said, “Abolitionists came to Jefferson County from the 1840s onward, but they didn’t reach a majority until Lincoln’s close election victory here in 1860.”

There’s really no way, according ChatGPT’s creators, to fix this issue because although they attempt to feed it reputable sources, it’s not currently possible to establish the correctness of all the data they use to train it. Also, the text data they used to train it ends in 2021, so ChatGPT can’t say much about anything more recent.

Still, it’s remarkable. So the next time you’re experiencing a challenging day, invite it to write a poem.

Find column archives at