GPT-3 doesn’t stop. It is the most productive text machine in history. And, among the many analyzes of its use that are now carried out, there are issues that do not cease to surprise, for example, the servility bias of the star artificial intelligence. An example is the tendency of ChatGPT to agree with the interlocutor even when he is wrong.
Since its release on November 30, 2022 by OpenAI, ChatGPT has received a lot of attention due to its ability to answer complex questions. Just five days after its launch, it had registered one million users. Now, a Google search returns more than 700 million results. These numbers will likely continue to grow over time.
ChatGPT is an advanced bot specialized in human dialogue. It is one of the latest examples of technology application Generative Pre-trained Transformer (GPT). In particular, ChatGPT is based on GPT-3, which was trained on more than 45 terabytes of data from different internet sources.
Offensive and violent content censored
GPT-3 is designed to mimic human speech and speech. Also, for security reasons, ChatGPT knowledge is limited until 2021 and strives to avoid generating offensive or violent content. In summary, ChatGPT has been designed to automatically complete the text, and, thanks to the information used in its training, it has additional abilities.
For example, being able to correctly complete the sentence “nine times three equals…”, we can say that he also has mathematical abilities; by being able to completeThe word Cheese, in Spanish, means…”, we can say that he has translation skills. Also, if you complete “Once upon a time, a professor from the University of Deusto was on his way to teach his Data Science class, when suddenly…”, we can say that he has creativity. These are just a few examples of the possibilities of ChatGPT.
Thanks to this paradigm, and the large volume of information used in its training, there have been many areas in which ChatGPT has shown capabilities comparable to experts. He has passed the exams to be a doctor in the United States, even being an author in scientific publications. This has generated controversy, and its use has been prohibited in schools, its texts have been prohibited on the popular platform stackoverflowas well as at scientific conferences.
Does not guarantee the veracity of the information
The system is trained to complete the given texts with plausible answers, as well as syntactically and semantically correct ones. This does not guarantee the veracity of the information provided or the coherence of the generated text.
We can say that, in some situations, he has a tendency to make things up, or to “hallucinate”, as professionals in the sector call it. Which leads ChatGPT to make mistakes on seemingly simple topics.
The lack of context may be one of the causes of your wrong answers. For example, if you are asked to complete the list “{Jose Maria, Jose Luis, Mariano}”, can suggest the name Pedro, correct in the context of the presidents of the Government of Spain. But it could suggest another name, like Enrique, in the context of Spanish names. Someone could even complete it with the names of neighbors, colleagues or friends if those names were present in their context.
Since it generates the text based on patterns learned from the data, it lacks a model of the world that allows relating concepts of the physical and social world. So, it might fail when answering questions like “What colour is the White horse of Santiago?”.
Other errors could be due to the poor quality of some of the texts included in your learning source, which can lead to unwanted biases.
Concerns over the use of GPT-3
ChatGPT is a type of Artificial Intelligence called Generative. These AIs are very popular these days because of how they can generate text, images, audio, and video in a similar way to how a human would. Different professionals can use them to generate content in an efficient and scalable way.
However, there are concerns, such as its potential to generate massive fake news or hate speech. There is also debate and discussion around the alleged violation of the intellectual property of the works used in their training, or the authorship and ownership of the works generated with these tools.
Some people believe that AIs can replace writers and other professions where creativity is important, such as design or music. Others see in them a useful tool to improve creativity and productivity.
The truth is that AIs are changing the way we think about writing, creativity and its potential. In the future, it is likely that these artificial intelligences will have a greater capacity to generate personalized content that is more accurate, truthful and adapted to different contexts. In addition, multimodal applications will emerge, which will be able to combine text, image, audio and video generation capabilities.
GPT-3 is already in our lives and it’s just a beginning. Let us take advantage of its potential, consider its limitations and do not ignore its risks.
