GPT Chatbot: helper, competitor or spoiler?
- February 20, 2023
- Posted by: Yury Subachev, PhD
- Categories: Paper writing, Useful tools
GPT Chatbot with artificial intelligence is today a record-breaker in terms of visitor growth. At the beginning of February 2023, its audience reached 100 million people. The programme was developed by an American company Open AL and was launched on 30 November, 2022.
The Chatbot is based on one of the most advanced language models, GPT-3 (Generative Pretrained Transformer 3). The scope of data used to train the model is estimated at hundreds of billions of words.
The popularity of the programme is explained by its extremely wide range of features. It allows:
- to write a code, translate it into another programming language, look for bugs;
- answer difficult questions, including giving medical advice;
- create texts: compositions, essays, scripts. Chat GPT can generate new episodes of TV series, set tasks for heroes of imaginary worlds and even create games based on a particular plot;
- write song lyrics and sheet music; write texts for scientific articles;
- develop a variety of plans, such as individual nutrition plan or energy consumption plan;
- translate from one language to another. Chat GPT knows more than 95 languages;
- generate financial analysis, summaries of scientific and technical articles and scholarly concepts.
The dialogue mode allows Chat GPT to answer additional questions, acknowledge your mistakes and make corrections to a written text.
Interaction with the programme takes place in natural languages. The Chatbot responds best in English, but other languages are also available, including Russian. Owing to this, any user can work with the programme, even those having no programming skills.
2. Disadvantages of the Chat GPT.
Of course, the Chatbot is not perfect. It has certain limitations, for instance
- It may give plausible but incorrect or meaningless answers. This is due to several factors. First, accurate information sources may not have been available during the training of the model. Another reason is that the programme is trained to reject some questions to which it could give a correct answer. Finally, supervised learning can lead to errors because the ideal answer depends on what the model knows, not the individual.
- The Chatbot is sensitive to wording of an incoming query. If the model responds that it does not know the answer a slight change in wording may help. Its answers can as well be corrected with suggestive questions.
- If an incoming query is ambiguous the model does not ask any clarifying questions, but simply acts at random.
- The creators have trained the model to reject invalid queries. However, sometimes it may react to malicious instructions or show bias. The model is not always good at distinguishing between dangerous and safe content. Errors can be one way or another.
- The artificial intelligence has not so far any information about the world after 2021. Questions concerning the subsequent period cannot be answered.
The developers claim that the Chat GPT is significantly safer and more efficient than its predecessors, GPT-3 and Codex. The training based on human feedback has significantly reduced the number of incorrect and unconducive results.
3. Chat GPT changes the reality.
The fact that the Chat GPT generates texts that could be written by humans has sparked active public debate. Many experts have raised concerns about the potential use of the software for writing educational texts, malware, phishing emails and automatic comments. Access to the Chat GPT on New York City public school computers is restricted because of its negative impact on education.
The education system specialists also have questions about the Chatbot. A number of cases are already known when students presented their graduation papers written with its help. For example, one of the universities in Russia demanded that such a diploma be revoked and that the use of the programme in Russia be banned.
The Chat GPT seems likely to change many familiar processes. In particular, the educational system will have to contemplate on what the traditional essay or composition can be replaced with. The fate of graduation and term papers is also unclear.
This raises a number of issues of ethical, economic and legal nature:
- It is obviously impossible to check the origin of essays, compositions and other educational texts. A way out of this situation could be the use of the old soviet system: the essay to be written in the classroom and the topics to be offered to the students just before they start working.
- The situation is more complicated with term essays, graduation papers, theses, Master degree theses and PhD dissertations. There is nothing to prevent the use of this or similar programme for writing scientific articles and even monographs. It is impossible to place the authors of all these texts in the environment isolated from the Internet.
- It is no secret that a huge number of university graduation papers, term essays and even dissertations are written by copywriters. This is a huge market. In this case, is there a big difference between natural and artificial intelligence if a student is not in fact the author of the work? Moreover, the real author may not inform the customer that he/she involved a computer programme to implement the work.
- Obviously, prohibiting the use of the Chatbot does not make sense at all, given that it is impossible to prove the use of artificial intelligence. If a tool that facilitates and speeds up the process has emerged it will obviously be used.
Then what is there to do? – To do what has always been done. The defence of diplomas, student’s essays and other papers is practiced to make sure that a person has not just written a text but has delved into the essence of the issue. So if one really knows the subject, answers the questions with convincing arguments, demonstrates the knowledge of sources mentioned in the bibliography, then the goal of the work has been achieved.
4. Chat GPT in science.
Every sound scholar realises that the most important thing in a scientific article is the research itself, its novelty, relevance, evidentiality. The text is secondary to the above, although its quality is of course very important. Possibly, if researchers pay more time and attention to the work itself and less to its description science will just benefit.
We do not know how some areas, such as philosophy or theology, will develop. It is probably a matter for philosophers to discuss.
What should a scientist who decides to use the Chat GPT take into account? – Of course, the disadvantages mentioned above.
- Anything written by the Chatbot should be treated critically and checked “with a magnifying glass” because the ability of artificial intelligence to generate scientific nonsense is just as good as that of humans.
- Facts, dates, quotes also need verification. The story of how Christopher Columbus set foot in America in 2015 is common knowledge. No one is immune to mistakes, including computer software. Blunders that are not deemed as errors by artificial intelligence may amuse people, but no self-respecting journal would accept such an article for publication.
- It makes sense to repeat the queries by changing the wording. If the Chatbot gives contradictory answers in this situation it is up to the human to deal with the inconsistencies.
Thus, one might say that while Chat GPT is more advanced than its predecessors it only helps the individual, but not replaces a human.
Now let us turn to practice.
5. How do I register and use the Chat GPT?
The use of the Chat GPT is free of charge.
Currently Open AL services are unavailable in the Russian Federation, China, Belarus, Afghanistan, Venezuela, Iran and Ukraine, but this prohibition can easily be circumvented with a VPN.
If you do not yet have a VPN, you can use free VPN extensions for Google Chrome. Choose a server location in a country where Chat GPT is available.
To activate an account in Open AL, a telephone number is required.
Proceed as follows:
- Switch on the VPN and select a country that has access to Open AL services. Press the “start” button.
- Go to the OpenAL website, click the “sign in” to go to the registration page and fill in the form. You will receive an email confirming your registration.
- Now you need to enter the phone number. Go to sms-activate, click Open AL on the left panel. You get number, now copy it and paste to Open AL.
- A sms code will appear on the sms-activate page, you need to paste it into a special box on the OpenAL registration page. After that, click on the orange checkbox.
Sometimes registration is closed. In such cases you should simply return to the website some time later and try again.
Once the registration is complete, a workspace opens. Type your query into the entry line and get the answer. You can ask clarifying questions if you are not entirely satisfied with the answer. As practice shows, the Chatbot will also answer to clarifying questions like “rewrite” or “correct”.
Conclusion
Can one be sure, given the technological progress, that a day will come when everything human beings do can be outsourced to machines? Will scientists be sunbathing on the beach playing “Preferans” while artificial intelligence is solving the problems of cancer and global warming?
I do not think so. The programme has reached its specific level of perfection because it is coached by a human being. But it is just a man! The entire tremendous progress of modern science has become possible due to the fact that lots of people with different experience, mentality and culture work over it. Even mistakes made by different people go to the treasury of science.
It is possible that the imperfection of multiple human brains in the aggregate produces in fact the breakthrough effect we are having today. Computer intelligence is certainly very powerful and we are still far behind it, but it represents only single-type intelligence and one style of thinking.
So, dear colleagues, remain on edge. The tasks we have to handle demand a coping period of more than a century.
Good luck with your scientific publications!
