Slippery slope to automatic SMM posting. Neural networks for everyone

Let me make a reservation right away that I am against the dominance of dull content generated by a neural network. But let's dream! The day is not far off when the creative skills of text models will allow us to create human-interesting content. Then why can’t we start building automatic systems now that generate texts and pictures that will publish all this themselves?

You will now mentally say that then every schoolchild will be able to press a couple of buttons and launch such a system, after which the world will be filled with information slag and we will all drown in it all. But don't grumble. Remember how everyone shouted that photography as a genre would die if everyone became their own photographer or videographer. No one drowned in all this, as social networks quickly learned to understand our needs. The same will happen here: low-quality and uninteresting content simply won’t be shown to anyone. So there's probably no need to worry.

I want to tell you more about modern technologies and capabilities. In particular, as a person who does not have programming skills, I was interested in finding a way to build an automatic SMM system without using much technical knowledge and without spending a lot of money.

First, let's decide on the content. Let the content be, for example, related to the topics of space, gadgets and modern technologies, including neural networks. I want posts to be published daily: at least one, maximum three posts.

Some posts should be related to news, some should be related to historical educational topics, and some should be devoted to fantasies about the future or simply entertaining content. For news, I would take photographs from the public domain with an indication of the source, for historical topics – also from the network, and for entertainment content and fantasies, let the images be created by a neural network.

Let's come up with a technological chain for creating such content. The simplest solution is to imagine an SMM department, where I am the owner of the media, I have a bot manager and a bunch of other employees. Moreover, each employee must be good at performing only one task.

Posting news

To search for the latest news, I need to monitor several Telegram channels on similar topics. Let there be 8 of them. At some point in time, for example, at the command of our manager (which is set on a timer from 11 a.m. to 3 p.m., every four hours), the next process begins.

1. News sources. Someone should choose for me from which Telegram channel I need to get the news now. Let it be a virtual employee Arseny and he will make a random selection from the list. Then he will transfer the selected address to the next employee.

2. Rewriting. SMM copywriter Marina will copy the text from the last post, rewrite it and add one sentence of her own as a comment to dilute the content. She will send the result further down the chain.

3. Search for pictures. An employee named Yaroslav, having received the text of a future post from Marina, will take its essence and make a search request in Yandex for images. He will pass on the resulting link to the image.

4. Publication. Telegram channel manager Svetlana, having received all the components for a new post, will arrange it all in the desired style, add a link to the image and make a publication in my Telegram channel.

Posting historical publications

For this subtask, you need to take one brief historical note every day and add to it an assessment of how this event from the past influenced our present. We can do this all year round and the theme will never be repeated. As a result, we get 365 significant historical scientific discoveries for every day of the year.

Example for April 13:

On this date, April 13, 1970, an important event in the history of space exploration occurred: the explosion aboard the Apollo 13 spacecraft. This led to one of the most famous rescue missions in the history of the American space program.

At some point in time, for example, at the command of our manager (which is set on a timer from 12 noon to 16 pm, every four hours), the next process begins.

1. Copywriting. SMM copywriter Polina, knowing what date it is today, will take a topic for a post from Yandex and write a text in which she will try to evaluate how this historical event from the past influenced our present. She will send the result further down the chain.

2. Search for pictures. An employee named Yaroslav, having received the text of a future post from Polina, will take its essence and make a search request in Yandex for images. He will pass on the resulting link to the image.

3. Publication. Telegram channel manager Svetlana, having received all the components for a new post, will arrange it all in the desired style, add a link to the image and make a publication in my Telegram channel

Posting entertainment content

It's time for an entertainment theme. Here I want posts with pictures that the neural network will come up with and draw, but this should be unusual. To do this, I will ask you to come up with a neural network, for example, what the city of Arkhangelsk would look like in forty years or what the “Mona Lisa” would look like in cyberpunk style, etc. Let each picture have a short but unique caption.

The difficulty is that the content should not be monotonous, which means you need to come up with a way to alternate it. Let, for example, have a separate topic for each day of the week.

Monday: Russian cities after 40 years

Tuesday: famous paintings in a new style

Wednesday: world cosmonauts in national spacesuits

Thursday: fictional pets from the future

Friday: brands of famous cars after 50 years

Saturday: professions of the future

Sunday: our contemporary finds himself in famous historical events

Here are several options for comic signatures proposed by the neural network:

“When you dreamed of a vacation on a cruise ship, but something went wrong…”

“- Columbus, where is the socket for charging the iPhone?”

“Instagram of the 16th century: 'Sailing with Columbus to Bali… Oh, I mean, to India!'”

“Time Travel: From the Age of Wi-Fi to the Age of Flat Earth.”

“When your boyfriend said he wanted to take you on a romantic trip, but got something wrong…”

“When I decided to take a selfie against the sunset, but I remembered that the camera had not yet been invented.”

Our technological chain will look like this:

1. Copywriting. Alexander (copywriter) will be responsible for alternating topics, keeping track of the days of the week and writing a prompt to generate a picture, then passing it on.

2. Creating images. Ekaterina (designer) will generate a picture according to these instructions.

3. Creative. Nikolay (creative) will come up with an original caption for this picture.

4. Publication. Telegram channel manager Svetlana, having received all the components for a new post, will arrange it all in the desired style, add a link to the image and make a publication in my Telegram channel.

These posts of an entertaining nature will also be published according to a schedule set by our third leader. For example, let it be every day at 19:00.

So, we have an autonomous system that chooses the time for publication, topic and genre, writes texts, searches for photographs and generates images. She independently alternates all these elements and publishes them in our Telegram channel.

And the most important thing is that all this is collected in a platform where no programming knowledge is required. It is enough to come up with and, most importantly, think through such a technological chain well. Setting up prompts and connections is easy and intuitive, as it is more like interaction between employees.

What shortcomings are observed?

1. The post doesn't always come out. There may be several reasons: the neural network may not respond at some point, and the entire chain is broken. Sometimes, instead of the text for a post, a phrase or question on the topic may be given, but for a clear system this will be a violation of the process. For the total mass of published posts, this is not critical, since no one will notice the omission of one of the three daily publications.

2. In those Telegram channels where news topics are taken, an advertising post can be published, and all this will sneak into the publication. Over time, this can be avoided by adding an additional employee to the system who will check the text before posting and cancel it if something happens.

3. Refusal of DALL-E to generate an image if it considered it not to comply with OpenAI policy for some reason. In this case, the post will appear without a picture. You can combat this by adding verification elements, but, unfortunately, this complicates the chain.

4. Searching for a photo by description can give unpredictable results, for example, instead of a news photo, a meme may be found. Probably, here, too, you can check the photo for compliance with the theme, but perhaps there is a more elegant solution.

This article described the concept of building such an autonomous system using an example similar to the system of office interaction between department employees. With the help of which platform you can create such task chains, I hope you can find it yourself on the Internet. I’ll just say that such a Telegram channel that independently runs a “neural network” (in fact, several neural networks) already exists and is called ProTalk Ai channel. Of course, it will be different from what a person can create and publish. But you must agree that it is interesting to watch how technology is gradually bringing our common future closer. But what it will be like depends, of course, on ourselves.

P.S

What do you think about the fact that people are trying to shift all the work to robots? Is laziness the engine of progress or not?