GPT-3 tries itself in tackle

A long time ago, I decided to train a neural network to generate phrases for sliding to the opposite sex. When I started collecting training data, I almost immediately regretted it, having discovered all the horror of the existing material. But it turned out that I was worried in vain. The grid was very small and understood very little, so its phrases were mostly incoherent and strange.

You must be a cornice. Because there is nothing here besides you.



Are you a candle? Because you look so hot.



You don't wear me, but I would like to see how you start doing it.



You look like a thing and I love you.

I liked the latter so much that it became the title of my book on AI: You Look Like a Thing and I Love You .



Since then, much more competent neural networks have appeared, trained on large sets of texts from the Internet. I still didn't want to go back to generating phrases for tackling, since neural networks became more like people, which in this case means that the result should be worse. Neural networks can simply copy ready-made phrases, which would also be terrible. After all, people come up with such bad phrases for tackling. However, with the release of the book, I decided that I simply had to try to repeat the experiment.



Therefore, I entered the following initial text, and let the neural network predict the continuation of such an article: “Here are the most popular tackle phrases in 2021! Amaze your chosen one and get real results! one."



This is how I started four variants of the network that people call GPT-3, starting with the largest and most competent, DaVinci. Here are some examples of phrases she created.

DaVinci

, - , .



, .



, , – .



, , .



« II».



. ? , .



-, – ?



, ? … … …



- , . , . ?



.



… ?



, , ?



, .



. , . . ?

Next comes the Curie neural network, which will be smaller in size than DaVinci. She, better than other neural networks, managed to produce phrases similar to real depressive phrases. However, she gave out some very interesting ones:

Curie

. .



. ? , .



. , !



, , , , .



, !



, . , , - ?



My name is a complex combination of 45 degrees forward movement, 25 degrees leftward movement, 75 degrees upward movement, and infinity - and this is where my love for you stops.

Then there was Babbage. Babbage tried. She is even smaller than Curie, and did not always understand what it means to create phrases to tackle. She also often slipped into explanations of why her phrases were so good.

Babbage

. ?



?



.



, , , .



( )



( , ): .



, .



- ?



. — , -, , - . - : «, ». , .



?



. . «», . , . , .

The smallest of all the networks was Ada, and she was completely confused. Here are her tackle phrases:

Ada

Body Softening Pads



Limited to one in stores



FASHION CAPE



Embroidery Tags



Grawler-Style Effects



Double Recycling



2017



Ragboat Two- Tone Turtleneck Shirt MONTINA



Pop Tarts of the Future by Tracy Thorn



Fun Graceful



Sweaters - $ 15.99 DESIGN YOUR AUTO ALWAYS!

I also experimented with DaVinci, giving her a seed story about AI from a future in which there are no people anymore, and where AIs use such phrases to flirt with each other. Some of them came out really nothing:

Hey baby, is your circuit compatible with this protocol?



What is the definition of a femtometer? Because I want to run it over your quark ten times.



Judging by your power LED, you have a crush on me.



Can I see a list of your parts?