Artificial Intelligence (AI) is the future of technology. Programmers have used this technology for the most diverse activities, from the most complex to the simplest, such as a simple chat. This, incidentally, is the premise of OpenAI, which created an AI chat interface that is being challenged by millions of people.
Read more: Meta Technology: artificial intelligence writes academic articles and TCCs
see more
Alert: THIS poisonous plant landed a young man in the hospital
Google develops AI tool to help journalists in…
ChatGPT is a GPT-based AI chat interface. According to Ars Technica, it is capable of writing poems, correcting coding errors, generating art from Artificial Intelligence – more or less like the app Lensa – and explain complex concepts.
The programmers managed to develop a special prompt that uses the Human Feedback Reinforcement Learning (RLHF) technique. This makes the AI based on how the humans using it rate their responses.
Just productive conversation
OpenAI wants this to be a chat for
conversations healthy, like a “virtual friend”. ChatGPT is programmed to limit and even not respond to harmful subjects and terms involving hate speech. This is the platform's biggest challenge.What happens is that it is difficult to control the offensive tendencies. Despite all the programming behind the chat, some people have already managed to bypass some filters of content asking, for example, for the Artificial Intelligence to look for exits in scenarios fictitious. Furthermore, the interface proved to be vulnerable to direct attacks.
fanfiqueira
Another problem that the interface presents is that it has a tendency to be “fanfiction”. You know that friend of yours who can to lie so well and with so much detail that everyone ends up falling into the story? Well, ChatGPT seems to have this same impulse.
According to OpenAI, the platform is good at inventing things that don't exist. For example, chat can create a book and give millions of details about it. As a result, the reliability of the tool is impaired.
But luckily, technology is constantly advancing. This is already an improved version of previous models. A possible GPT-4 would be under development and, if used in ChatGPT, could improve the tool even more.
Would you be able to challenge this AI chat?
As it is still in the testing phase, the chat needs people to challenge it. To do this, simply create an account on the OpenAI website – it's free and fast. Afterwards, just enter the ChatGPT website by clicking here.
Unfortunately, for now the platform is only available in English.
Graduated in Social Communication at the Federal University of Goiás. Passionate about digital media, pop culture, technology, politics and psychoanalysis.
