Copy.ai’s AI-based article creation system is astonishingly “usable” and supports Japanese | GLM

Copy.ai’s AI-based article creation system is astonishingly “usable” and supports Japanese

TechCrunch mentioned earlier that Copy.ai, an AI-based writing service, has successfully completed a funding round. The startup uses GPT-3 to support enterprise users’ article and writing projects. GPT-3 is an acronym for Generative Pre-trained Transformer 3, an AI system developed by OpenAI. It can receive text from the user and generate a great variety of sentences on behalf of the user.

When I covered Copy.ai’s funding round, I actually tried writing with AI. As a result, I found it to be a very attractive idea. I haven’t had the unspeakable fear of computer-based automation of writing, but rather I was curious. When I interviewed, the Copy team insisted that “GPT-3 is a great AI writing tool,” so I was interested.

To satisfy this curiosity, it is quick to actually use it. So this morning, I compared it with Copy’s rival Headlime. In conclusion, I got the impression that the products that Copy’s team built on GPT-3 technology have great capabilities.

Of course, GPT-3 can expand the provided materials to create articles. But this technology can do more than that. The Headlime app with GPT-3 was able to write articles that could be evaluated as moderately good. Not only that, it expressed a concept that was in my head when I interviewed the material, but was not touched on by the material itself.

This is a subtle part and difficult to explain well. You’ll have to see an example of how GTP-3 made its own decisions when using Headlime for the first time. I deliberately provided only the facts of the material about the current stock market trends and tried how the system handles it.

Screen Shot 2021 03 17 at 10.14.51 AM

As a bipedal animal that simply sticks meat to a stick with a high calcium content, I wondered if I could prove to be intellectually overwhelmingly superior to a computer made of silicon. As a matter of fact, I found that the material information required at least 500 characters, so I slammed a little to fill the specified length.

Screen Shot 2021 03 17 at 10.16.31 AM

Now it’s the next step. At the push of a button, I saw some suggestions for the title of the article I was asking for. To be honest, they were all pretty accurate.

Screen Shot 2021 03 17 at 10.17.30 AM

Around this time, I began to wonder, “Isn’t this good?”

I liked the headline because I thought the first candidate was good, so I chose it and proceeded with the work. Next is the task of putting together the lead part of the article. It’s usually a brain-intensive task, in this case just a click.

Screen Shot 2021 03 17 at 10.17.47 AM

Another option presented here is:

Screen Shot 2021 03 17 at 10.18.19 AM

This wasn’t bad either.

What impressed me was that each option was not just a change in the details of the wording. Focusing on the different parts of the rather sloppy 500 characters I typed into the system, the sentences were different from the syntactic level. I felt that GPT-3 should have been given a slightly better material. The system was working very hard, but I wasn’t.

2021 03 18 Screen Shot 2021 03 17 at 10.22.48 AM 1
After fixing the lead part, some kind of CMS-like thing appeared, and it was required to write the text after the lead part. I was already willing to take on the humble servant of the system, but soon the system offered to stop writing and take over.

2021 03 18 Screen Shot 2021 03 17 at 10.23.50 AM

Since it was a free trial, I couldn’t judge at all how much processing 2 credits was worth, but anyway, I pressed the button called Write for me. The result is the next article.

2021 03 18 Screen Shot 2021 03 17 at 10.24.17 AMI have entered confusing sentences including dashes as material, but let’s see how this software handled it. It seemed difficult to dash, but he quickly returned and argued correctly that “rising interest rates would make non-traditional, high-risk investment classes like venture capital unattractive.” I was astonished.

I will continue to experiment with GPT-3 technology and products based on it from different angles. Please look forward to the report. But now I want to stop and share the joy of discovering new and promising technologies. To be honest, this is amazing. It’s been a while since I was surprised by the new technology. GPT-3, well done. It’s amazing.

Rachel Maga
Rachel Maga is a technology journalist currently working at Globe Live Media agency. She has been in the Technology Journalism field for over 5 years now. Her life's biggest milestone is the inside tour of Tesla Industries, which was gifted to her by the legend Elon Musk himself.