WHAT IT IS‍

ChatGPT (an acronym that stands for Generative Pretrained Transformer) is an artificial intelligence model developed by OpenAI, based on human-machine interaction. It is a chat in its own right, conversationally imprinted and aimed at solving multiple questions. The special feature, as well as the extent of its introduction, lies in the fact that the platform learns through the questions it is asked.

All this is possible through the use of Reinforcement Learning from Human Feedback (RLHF), a method that allows this model to improve its responsiveness, both in terms of accuracy and with respect to the range of subjects in relation to which it can concretely generate solutions. Relying on the comparison of available data, ChatGPT is not designed to predict future events, but rather to solve contingent situations; in this sense, those who need to write an article on a given subject, as well as a programmer in order to obtain the correct code string, can benefit from it.

Indeed, the user input turns out to be crucial: the clearer and more specific it is, the better the feedback, hence the output. At the moment ChatGPT does not allow intervention in relation to sensitive topics, partly because, by OpenAI's own admission, this model is capable of producing potentially "biased and dangerous" responses. Scenario, the latter, which the company aims to limit as much as possible, contextually to the obvious implementation of elements that improve its use.

‍

HOW IT WORKS.

The wording says all you need to understand the ways through which to interact with the platform. It is indeed a chat in its own right, only on the other side there is no longer a simple bot, trained to answer specific questions in a mostly generic way. ChatGPT tailors the answer based on the ability, if you will, to ask the question.

As already pointed out, this model is designed to collect and subsequently process a substantial amount of data, which enables it to refine the accuracy and relevance of the response, calibrating it in as detailed and truthful a manner as possible to the needs of the individual user. In fact, what is established with this type of artificial intelligence is a real dialogue.

The latest version, GPT-4, launched on March 14, 2023, introduced the ability to input not only textual prompts but also visual prompts, such as photographs or images.

‍

RISKS AND BENEFITS

Given the novelty of this tool, at least in terms of the degree of sophistication so far, one can at best look out as much on the risks as on the benefits, all of which are potential at the moment.

In both cases, the ramifications are many, and the impact of such a phenomenon can only have repercussions at multiple levels. For example, recently the Garante della Privacy recently blocked ChatGPT in Italy for unlawful collection of personal data and lack of systems for verifying the age of minors, the first case in the West, a problem that has since been recessed in the face of assurances provided to the entity by OpenAI.

On other fronts, there is the possibility that the use of such a platform will make life even more difficult for professional writers, especially when it comes to content such as may be related to a neutral type of information (e.g., the promotion of an event). If in fact the work of the developers is on the one hand aimed at making the model capable of producing the most truthful feedback, there is also an improvement as well with regard to the variety of responses, which in fact certifies the ability to offer multiple types (or styles) for the same content; as already mentioned, the assumption is that the most suitable command is given.

Legal and professional aspects go hand in hand with ethical ones, a subject that represents perhaps the most complex challenge facing such a platform. The phenomenon has so far been curbed by precluding a priori the opportunity to move into the realm of controversial and sensitive matters.

However, with the intensification of its use, and the consequent, possible growth of its status as a reliable or at least useful tool, ChatGPT could acquire a somewhat uncomfortable credibility, exposing it to criticism due to its ultimately all-encompassing approach. On the other hand, we are talking about an artificial intelligence with which it is possible to converse in order to get answers, so it is good that first and foremost the user learns to relate to it in the most sensible as well as healthy way possible.