Chat GPT Detector

Via The Passive Voice, this link to a detector that will tell you whether the text you put in was generated by a real live human or by AI.

Obviously I couldn’t resist trying that out. Was every line of TUYO written by a real human person?

It didn’t seem reasonable to put in the whole thing, obviously, but I dropped in the first six hundred words. What is the result?

Your text is Human written

Good to know! Every single line? No! “ Highlighted text is suspected to be most likely generated by AI* That’s six percent, roughly. What are the suspect lines? Here they are:

***

Beside the coals of the dying fire, within the trampled borders of our abandoned camp, surrounded by the great forest of the winter country, I waited for a terrible death. I had been waiting since midday. Before long, dusk would fold itself across the land. The Lau must surely come soon. I faced south, so that my death would not ride up behind me on his tall horse and see my back and think that I was afraid to face him. Also, I did not want to look north because I did not want to see that trodden snow and remember my brother leaving me behind. That might have been a different kind of cowardice. But I could only face one direction. So I faced south. The fire burned low. My brother had built it up with his own hands before he led our defeated warriors away. Now it was only embers, and the cold pressed against my back. I wished I could build the fire up again. Mostly that was what I thought about. That was as close to thinking about nothing as I could come. It was better than thinking about the Lau. I hoped they came before the fire burned out, or I might freeze to death before they found me. Even an Ugaro will die of the cold eventually, without fire or shelter. I tried not to hope that I would die before they found me. Then I heard them, the hoofbeats of their horses, and there was no more time for hope. I held very still, though stillness would not protect me now. Nothing would protect me. I was not here to be protected. They came riding between the great spruces and firs, tall dark men on tall dark horses, with the Sun device of their banner snapping overhead in the wind. Ten, twenty. Twice twenty. And even this was only the vanguard. I stood up to meet them, raising my hands to show that I was bound to a stake driven into the frozen earth—to show that I was tuyo, left here for them. They looked at me, but they rode past, down the trail my brother and our warriors had left. They rode through the remnants of our camp, around the fire and around me, and a little distance more. At first I thought they meant to leave me to die alone in this place while they went on to pursue a broader vengeance against my people. That would have been a death even more terrible than the one a tuyo should face. But then they came back and circled around me, not many paces away, looking down at me. My relief was so great that for the time it pushed away fear. I knew immediately which must be their warleader. My people prefer silver, which is the metal that belongs to the Moon. The Lau decorate their warleaders with gold, as befits the people of the Sun. This man had gold thread worked into the collar of his coat and the backs of his gloves and the tops of his boots. He did not carry a sword or any weapon, only a polished black stick as long as a man’s arm, with gold wire spiraling around its length. I had seen illustrations, so after a moment of puzzlement I recognized this as a scepter. This man was not only a warleader, but a scepter-holder, carrying the authority of the summer king. I had not known any such had come to the borderlands. At least my death would come at the hands of a worthy enemy. The scepter-holder’s horse was the color the Lau call fire bay and we call blood bay, which is common for their animals and very rare for ours. It was a fine animal. The Lau breed beautiful horses, but they belong to the summer country. They are too long-legged and too thin-skinned for the cold of Ugaro lands. Like their horses, the Lau are long-legged and thin-skinned, and they like the cold no better. They are a graceful people, with elegant features and smooth brown skin. Lau men often grow beards, rare for Ugaro men, but they shave them short, just to outline the jaw and mouth. The warleader had a beard like that. He had cut his hair short to match. No Ugaro man would do such a thing; for us, cropped hair is a mark of shame. We tie our hair back or leave it loose, but we do not cut it. For a moment, while the warleader gazed down at me, the silence was almost complete. A horse picked up one foot and set it down again, and the wind blew across the snow, and leather creaked as a man shifted his weight in the saddle. Other than that, there was no sound. At last the warleader dismounted. He was far taller than I; even taller than most of his own people. He looked cruel to me, with a hard set to his mouth. I knelt and bowed my head to show the proper respect the one defeated owes to the victor. He looked at me and then at one of his people who had come up beside him. He said to that man, “We must have pressed them even harder than we knew, if they’ve left a tuyo for us. I suppose this must be the son of an important Ugaro lord, but he seems merely a boy.”

***

That’s interesting! Isn’t that interesting? What about those lines leaps out at the AI Detector? It’s a puzzle. “No more time for hope,” I suppose. “Other than that, there was no sound.” The other highlighted sentences are connected to those lines. Maybe that’s why the AI Detector is suspicious of those sentences.

Let me try a couple of paragraphs of an exceedingly boring report I’m working on today:

Your text is human written.

This is VERY human-written — only 2% suspected of being AI generated. The style is, you will see, very different. I’m a little surprised it seems human-written. I think of this as Bureaucratic Bullshit Style, which I would have thought would seem pretty generic and potentially AI generated. I guess not. I guess maybe it’s all about detecting phrases that other people have used? You can see the one highlighted sentence is extremely generic and has no doubt been used verbatim in roughly a hundred thousand papers and reports this century alone. Here are the paragraphs:

***

The EXCEL Program has now accumulated data regarding tutoring over 43 semesters. The data below include all active EXCEL participants who requested tutoring at least once during a semester for a class. These data exclude tutoring directed toward improving Accuplacer or other test scores. When a single participant requested tutoring for multiple classes, these classes are counted separately. Grades of C, B, and A constitute “success” in attempted classes. Other grades, including W or I, are considered to constitute lack of success in attempted classes.

The primary interest here regards the difference in success rates before Spring 20202 compared to success rates following that semester. Therefore, data has been summarized for the 37 semesters prior to Spring 2020 and the 5 semesters following Spring 2020. To allow a more granular perception of these data, the 37 semesters from Fall 01 to Fall 2019 have also been broken into sets of ten to seven semesters; Spring 2020 is then presented in isolation; and the five semesters following 2020 are then presented as a set. These data are summarized in Table 1.

Data regarding each separate semester are also presented in Tables 2-7, with semesters grouped into blocks to allow easier visualization of trends or the lack thereof.

For purposes of discussion, the “low tutoring” group here includes students who attended just one to three tutoring sessions during the entire semester. Categories are then nested, so that the category “4+” contains all students who attended four or more tutoring sessions, including all the students in the “7+” and “10+” categories. The “10+” category constitutes the “high tutoring” condition. The categories “4+” and “7+” are termed “medium-tutoring.” The greatest number of students attend few tutoring sessions, so sample sizes are low for the high-tutoring group during many semesters. Data regarding number of students requesting tutoring services are presented in Table 8.

***

As promised, exceedingly boring.

Also unfinished, so I’m going to go work on that now.