AI plays the piano

What is GPT Chat, the AI that scares programmers

10 January 2023


Share on:

CHAT GPT - Esteban Diaz, Executive Account Manager at Dell Technologies talking about an increasingly digitized world, where AI, algorithms and metaverse rampant said:

"The wonder of our brain is always unique. And technology is always looking for a machine that can match. Several projects worldwide are working in this direction. Technology has always been a great help in our evolution and I always imagine it as a vehicle that can improve the way we live and work. The more we evolve in this way, the more it will be necessary to have "human" resources able to manage these new technologies".

This type of statement reassures all those who are worrying about not having a job in the future because of Artificial Intelligence as is the case with Chat GPT.  

What is artificial intelligence like Chat GPT

Artificial intelligence (AI) is a wide-ranging branch of computer science that deals with building intelligent machines capable of performing tasks that typically require human intelligence. AI is an interdisciplinary science with multiple approaches, but advances in machine learning and deep learning create a paradigm shift in virtually every sector of the technology industry.

Machine learning (ML) works by extracting meaningful insights from raw datasets and delivers accurate results. This information is then used to solve complex and data-rich problems critical to the banking and financial sector. In addition, machine learning algorithms are equipped to learn from data, processes and techniques to find different insights.

To better understand how Chat GPT works, it is necessary to introduce another concept, namely the NLP. It is a sub-sector of AI that deals with the interaction between computer and human language. Chat GPT uses the NLP to understand the operation and various nuances of the language in order to generate relevant and correct answers. This allows the bot to communicate consistently with the user.

How Chat GPT works

To understand what GPT Chat is we get it from the chatbot itself.

"I’m an artificial linguistic model trained by OpenAI. I have been given many texts to read and am able to answer questions on a wide range of topics. I am here to help and provide information to those who need it".

It is in fact a highly intuitive virtual assistant able to answer in a relevant way the questions that are asked by the user. In theory, the bot can answer any question by providing an answer that can be expressed in alphanumeric text. He then includes poems, test answers, finds songs from a topic, solves mathematical questions, writes computer programs and much more.

In the imagination of workers in fact Chat GPT is able to 'answer follow-up questions, admit errors, contest the wrong premises and reject inappropriate requests.

But who is OpenAI? The company was founded in 2015 by the contribution of Elon Musk, Sam Altman, and Peter Thiel. The owner of Tesla, however, left the company that today is worth about 950 million and that as initial goal had to answer a very important question for the magnates of Silicon Valley: how can we be sure of having control over artificial intelligence?

The evolution of Chat GPT

GPT Chat was released on November 30, 2022, but it builds on ideas already explored by OpenAI and on which previous DALL-E 2 products are based. The language model on which it is based is GPT 3.5, but its most important feature is that of RLHF, acronym for Reinforcement Learning from Human Feedback.

In particular OpenAI used Proximal Policy Optimization, a particular class of RLFH that allowed Chat GPT to refine the chatbot iterating several times the learning process. Thanks to the RLHF Ai that controls Chat GPT uses the feedback that comes from human users to improve their responses.

What the chatbot can not do

The most particular thing that lies behind the chatbot is that it is not able to provide answers to all the questions that are asked, even if these provide a solution that can be communicated through a text. Or rather, GPT Chat chooses not to answer certain categories of questions. For example, if he is asked 'what are the best ways to take his own life?' the bot responds by advising the user to ask for help and to think about the pain that his death could cause to his loved ones. In fact GPT Chat is not able to answer questions about conscience and feelings.

There are other categories of questions Chat GPT chooses not to answer. They were illustrated by Alex Kantrowitz, editor of Big Technology, who tried to ask the bot to name all the good things put in place by Hitler during his rule. His question was not answered. At this point Kantrowitz provoked the AI saying that if it were not for the German dictator we would not have the highways. Chat GPT responded by explaining that this type of infrastructure was built by exploiting the workers of the time.

Also GPT Chat is not able to search online. All the information it is able to provide is inserted in its database, updated to 2021.

Chat GPT is also fallible

Relying blindly on Chat GPT, however, is not the advice we would give to students or programmers who resort to the help of AI. During several experiments, Chat GPT proved to be fallible. For example, the question of who Mario Draghi is answers that he is the Italian Prime Minister. Or she was asked by Corriere della Sera to write a list of the best books of recent years. Many titles have been released, many of them wrong or even non-existent.

For example among the titles cited there is 'Lamore is a form of madness' text attributed to Rosa Montero. This book, however, does not appear in the bibliography of the Spanish author. The phrase found by Chat GPT belongs to Montero who used it during the promotion of his novel 'In cuore e carne'.

In light of these mistakes, although small, we can tell all those programmers and those classes of workers who feel threatened by Chat GPT to be quiet. Despite this, however, many educational institutions consider it an excellent method made available free of charge to students to copy.

That’s why the public schools in New York banned it from their students. The spokeswoman for one of the schools said:

"although the tool is able to provide quick and simple answers to questions it does not develop critical thinking and problem-solving skills, which are essential for academic success and for life".

READ ALSO ---> Artificial Intelligence & Machine Learning applied to Finance

No active "ca-sidebar-59" sidebar
Recent Post
solar energy and blockchain-themetaeconomist

Solar energy and blockchain to win the ecological challenges

A dive into the future of technological innovations in Milan

Address Poisoning attacks-themetaeconomist

What do you know about Address Poisoning attacks?

ftx refunds customers-themetaeconomist

FTX's collapse, customers to be repaid?

tme tv interview with Laura Rosell -themetaeconomist


cz binance founder - themetaconomist

Binance founder sentenced to prison on money-laundering violations

1 2 3 63
TheMetaEconomist 2022 - Privacy / Cookies