AI and Machine Learning are Handy with Essay Writing
We don't even notice how artificial intelligence is firmly embedded in our lives. In many areas, its work is invisible. But soon, artificial intelligence can bring fundamental changes everywhere. Not only where processes can be automated, but also where creativity is needed. Thus, the Cheapwriting. services company suggested that AI will be able to perform some of their services instead of real authors. For example, it will cope with writing essays or term papers.
You have probably heard by now that in the future machine learning and robots could put people out of work. This could be a problem for cashiers, longshoremen, and other professions where physical labor is important. But artists, writers, and other creative types could also be at risk. In June 2020, OpenAI introduced the GPT-3 system, which is capable of writing text by following set instructions. And it is already doing a great job for journalists. The Guardian recently published an article written entirely by artificial intelligence.
AI is capable of many things, but it struggles to cope with the natural language with which humans communicate at ease. As it turns out, this statement is outdated: AI is capable of writing a full-fledged term paper in a matter of minutes.
The results of a new experiment proved that, with the right settings, AI can not only write but make the term paper hundreds of times faster than a human.
For the experiment, EduRef chose the GPT-3 (Generative Pre-Training Transformer 3) algorithm. This algorithm has more natural language skills than all existing analogs. Many believe that the main problem of AI is the lack of common sense. Instead of simple and logical phrases, the machine can use completely inappropriate words. The task was to teach the algorithm, which was entrusted to college graduates and teaching staff.
As a result of the experiment, the AI had to write a term paper. As it turned out, the GPT-3 failed in only one of the four subjects.
The researchers note that the machine was able to mimic the human style in the areas of grammar, syntax, and word frequency. Although, of course, there were too many technical phrases in the work. In general, AI received about the same reviews as ordinary people. This means that soon, to write a course paper, you will only need to configure AI in the right way, and not spend dozens of hours studying the materials.
Two sides of the force
One of the commercial applications of AI is generating unique content for websites. A research team from the Washington and Allen Universities has presented a state-of-the-art AI model named Grover. Its main task was to study and detect fake news.
Grover is a good disinformation detector. If fake news is generated en masse, it becomes even easier to detect - with over 97% accuracy. AI shows good results in detecting fake news written by different classes of generators.
According to scientists, fake news authors are not easily fooled by Grover. Still, certain topics are more difficult for AI - for example, news about finances.
The secret of Grover, the researchers write, is that it itself can act as a generator of fake news. This is because the AI is most familiar with the peculiarities of their writing. This is the use of common or predictable words - as well as the workings of such generators.
Moreover, if you specify a domain for Grover, it will generate text similar in style to articles from this site.
What is GPT-3 and how it differs from its predecessors
Generative Pre-trained Transformer 3 is currently the most complex language model. The main difference between this algorithm and other similar ones is its trainability. The system is trained on 1.5 trillion words, and its largest version takes up about 700 gigabytes.
GPT-3 generates text based on 175 billion parameters, a value that reflects its processing power. Depending on the number of parameters, the system evaluates the data better or worse and gives some of them a higher value and some a lower value.
The essence of the new algorithm has not changed compared to the previous version. A neural network analyzes huge amounts of data from the Internet and tries to predict the text word by word. But it still needs a reference point - a query to work with.
And the more input data you give the system and the more attempts it has, the more convincing the text can be. For example, if you give it the beginning of a famous poem in the style of one author, it can continue it in the style of another.
Design, music, stories - what the GPT-3 can already do
Many developers have already tested the algorithm in a variety of scenarios: from songwriting to creating code and musical arrangements. According to one of the developers, in most cases, the system gives a convincing result, if not from the first, then from the second or third attempt.
For the most part, scientists used the GPT-3 to generate ordinary text: stories, songs, press releases, technical docs. But one of the developers went further and asked the neural network to write a text about itself. There was an article with the title "OpenAI's GPT-3 could be the greatest phenomenon since Bitcoin." It is quite possible that GPT-3 will soon advise on Best Bitcoin Loans.
Other developers discovered that Pre-trained algoritm could generate any kind of textual information. This includes guitar tablature and computer code. For example, the Sharif Shamim developer has shown that the system can work with HTML markup instead of natural language and create a layout based on text queries. For instance, you can say what button and layout a website should have - and the neural network visualizes them.
Also, with GPT-3 Shamim has created a simple React-based application generator. You only need to write what the program should do: the algorithm itself will translate the request into uncomplicated code.
Another developer created a plugin for Figma based on GPT-3. It allows you to create a design by simply giving the neural network a text description.
In one example, an American student spent two weeks blogging about success and motivation generated by neuronet. Out of curiosity, he launched a blog promotion and received 26,000 visitors. From this amount, almost no one guessed that a smart algorithm wrote these texts.
The creators of the free AI Dungeon text-based quest game also accessed the GPT-3 and updated the app. They claim that after that the players got full freedom of action: the system responds correctly to all requests and comes up with a world based on them.
GPT-3 is not real AI, but it can seriously affect the world
In recent years, it has become customary to call artificial intelligence almost everything that concerns neural networks and machine learning algorithms. This is easier for many journalists and users who are not involved in development. However, in fact, humanity is still far from real AI, and in the neural network there is, in fact, no"intelligence".
Many early users of GPT-3 said that the algorithm writes text that is indistinguishable from human text and produces meaningful sentences. In fact, inside it is still the same text-on-demand generator — with all the advantages and disadvantages. Many people mistakenly believe that the system understands the context. In fact, it evaluates the connections between individual words and places the most likely words one after the other.
However, the disadvantages of Pre-trained algoritm do not affect the usefulness of the neural network as an application tool. Although the algorithm cannot yet replace humans, it can seriously simplify people's lives in general. Students will especially benefit. They will be able to relieve their daily academic responsibilities in terms of essay writing.
It is hard to say how far the implementation of the system will go. But it can be used in almost any activity: from teaching languages and programming to simplifying daily routine.