https://wordpress.org/

Chat GPT: Deciphering the new artificial intelligence tool everyone is talking about

Published on 25/01/2023
Partager sur :

Sometimes a source of concern and fear, sometimes a source of hope, artificial intelligence generally leaves no one indifferent. In recent weeks, one of them has attracted all eyes on her: ChatGPT.

If you’ve ever used a voice assistant like Siri or Alexa, you’ve already had a taste of what ChatGPT looks like. But unlike those voice assistants, ChatGPT is a real-time conversation tool that allows you to converse with it as if you were talking to a friend.

Let’s take a closer look at this new phenomenon that once again widens the field of possibilities offered by artificial intelligence. 

In this article, we start by clarifying some concepts. What are artificial intelligences? Their concrete applications, we also define how the ChatGPT tool fits into these definitions.

We then get to the heart of the matter and explain how ChatGPT actually works, what its limitations are and try to anticipate potential future improvements. We will then look at the best practices to be followed when using the tool and finally the different ways to get the most out of it..

Opening the Doors to AI: What’s behind Artificial Intelligence ?

Artificial intelligence (often written in its abbreviation AI) can be defined as the ability of a machine to display human-like capabilities such as reasoning, learning, planning and creativity.

AI enables technical systems to perceive their environment, deal with what they perceive, solve problems and act to achieve a specific goal. The computer receives data – already prepared or gathered through its own sensors such as a camera – processes it and responds.

Most of the time, AI systems are capable of adapting their behaviour to a certain degree by analysing the effects of previous actions and working autonomously.

Some AI technologies have been around for years, but advances in computing power, the availability of enormous quantities of data and new algorithms have led to major AI breakthroughs in recent years.

The types of AI:

The subject of artificial intelligence is widely documented and researched. In order to offer an understandable and concise vision of the subject, we will approach the subject by mentioning that there are 2 kinds of AI: Weak AI and Strong AI.

Weak AI


Also called Narrow AI or Artificial Narrow Intelligence (ANI), those technologies are used to perform simple tasks. This form of intelligence works through algorithms. The algorithms, programmed by humans, allow a robot to simulate intelligence while being assisted.

Thus, the so-called “weak” AI defines a program that simulates intelligence for the purpose of performing a specific task. In other words, weak artificial intelligence performs a predefined task. When it received the task, the machine analyzes it in order to solve the problem. This will then be transcribed into algorithms. The word “narrow AI” would thus be more accurate since this type of AI is anything but weak. It is the most common of the AI that surrounds us today. It enables some very robust applications, such as Apple’s Siri, Amazon’s Alexa, and autonomous vehicles.

Strong AI


In order to feel real feelings, the “strong AI” consists in making a machine understand what led it to perform actions. In this way, the system reproduces in an identical way the functioning of the human cognitive system. This type of AI refers to a model in which a machine would have an intelligence equal to (or at least very close to) that of humans. It would have a self-awareness that has the capacity to solve problems, to learn and to plan the future.

Thus, although deprived of natural consciousness, which remains specific to living organisms, the machine acquires an experience allowing it to modify its own functioning. In other words, a learning process results from the reception of basic data and leads to initially unprogrammed reactions. While strong AI is still entirely theoretical with no practical examples in use today, that doesn’t mean AI researchers aren’t also exploring its development.

Deep Learning and Machine Learning, let’s make it clearer:

Here are two more concepts that we encounter very regularly when we look at the question of artificial intelligences. First of all, we can affirm that deep learning and machine learning are sub-fields of artificial intelligence, and that deep learning is actually a sub-category of machine learning.

Machine Learning is an AI that can automatically adapt with minimal human interference, and Deep Learning is a subset of Machine Learning using neural networks to mimic the learning process of the human brain.

Several major differences separate these two concepts. Deep Learning requires larger volumes of training data, but learns from its own environment and mistakes.

On the contrary, Machine Learning allows training on smaller datasets, but requires more human intervention to learn and correct its errors.

In the case of Machine Learning, a human must intervene to label the data and indicate their characteristics. A Deep Learning system, on the other hand, attempts to learn these characteristics without human intervention.

To give a more concrete example, Machine learning is like a robot that follows instructions. It looks at patterns in the information to do things like predict what might happen next. Deep learning is like a robot that is smarter. It can figure out more complex patterns and make its own decisions about what might happen.

If we now go back to ChatGPT, it uses a form of deep learning called Transformer Networks. The Transformer Network is based on a technology called Attention Mechanism which enables the chatbot to understand the context of conversations and respond more accurately.

Concrete example of usage of A.I.:

The uses of A.I. are extremely wide and varied. Their applications therefore seem endless and full of promise in many sectors, expanding the field of possibilities and pointing to a bright future. The examples are numerous and the list is long, let us simply mention some examples in different sectors of activity.

In the healthcare sector, AI can be used to diagnose and treat medical problems, analyze patient data, provide personalized healthcare and automate administrative tasks…

Regarding the education sector,  A.I. are useful to provide personalized learning experiences, automate grading and optimize curriculum.

In the transportation sector, A.I. can be used to automate the transportation process, manage traffic and improve safety.

For manufacturing, AI can be used to automate the production process, improve product quality and optimize the supply chain.

In finance, A.I. automate investment portfolios, provide personalized financial advice, detect fraud and improve customer service.

In the retail industry : A.I. can be used to personalize product recommendations, automate inventory management and optimize the customer experience.

ChatGPT, deciphering of the new phenomenon.

ChatGPT has been a hit since its launch at the very end of November 2022. In less than a week, the platform already had more than a million registrations and hundreds of screenshots of conversations have gone viral on Twitter. Users have had a great time testing the tool, pushing it to its limits (which we will analyze a little further in this article), often showing great inventiveness and being surprised by the great power of this new tool.

ChatGPT, short for Chat Generative Pre-trained Transformer, is a natural language processing (NLP) technology that enables machines to understand and respond to human language. Developed by the American company OpenAI, it is a type of artificial intelligence (AI) system that utilizes deep learning algorithms to generate human-like conversations in natural language. In short, it is a form of text-based interaction between machines and humans.

As detailed earlier in this article, it is based on the Transformer model and uses large amounts of data to build a deep learning model that can understand and respond to natural language.

The model was trained using text databases from the internet. This included a whopping 570GB of data obtained from books, webtexts, articles,… present on the internet. To be even more exact, 300 billion words were fed into the system.

The platform is specifically designed to understand and generate natural language. This makes it different from many other chatbots, often designed to provide pre-determined responses to specific inputs or perform specific tasks. Its ability to remember the previous questions you asked and respond differently makes it stand out from all the other chatbots out there. The answers are also less “conversational” compared to those given by a more classical chatbot.

Although the web interface is in English, ChatGPT understands and responds in several languages including French, German, Russian and Japanese.

The end of the classical search engines ?

Its ability to answer questions, to paraphrase, to explain certain concepts in a simple way let some say that ChatGPT is able to make Google fall from its number one status. Indeed, according to some experts, AI text generators have the potential to overtake Google’s search engine due to their increasing proficiency in interpreting and responding to natural language queries. This can result in more precise and pertinent results than those provided by Google, which mainly depends on keyword searches. These tools will more than probably also improve with time as they are exposed to more data, gradually becoming better at capturing user needs and context.
It is still uncertain whether AI text generators will substitute search engines, but they are already influencing how information is located and experienced online.

How will this continue to grow ?

When looking at the tool itself, it is hard to predict exactly what the future holds for ChatGPT as it depends on many considerations, like technological progress and the aims of the OpenAI development team. It is likely, however, that it will keep on being improved through ongoing learning and the integration of new elements. There is a chance that it could be utilized more widely in the future to help organizations answer customer queries online, or to offer remote help to those who need it in everyday life. It could also be incorporated into fresh platforms or programs, allowing users to make use of it in even more comfortable ways. Furthermore, it is probable that ChatGPT will further refine its ability to recognize human language and generate more logical and pertinent replies. By utilizing newer machine learning techniques and training the model on fresh sets of data, it could possibly become even “smarter” and more adept at answering a wider range of inquiries and requests. There are thus many potential possibilities for ChatGPT to develop in the future, and it is exciting to witness the app’s ongoing development and improvement.

But ChatGPT is not alone in the field of tools that allow users to converse with an AI system and have it execute commands. We can mention other tools like Jasper, Paragraphai, or Rytr.

The developments observed with those technologies may lead us to believe that in the future this type of AI will become even more “intelligent”. For instance, instead of searching for a specific video on YouTube such as ” how to hit a tennis ball properly “, you could just input your goal, ” I want to improve my tennis ball touch ” and the system will provide you with a customized course.

Moving from content-based queries to higher order goal-based queries that can be summarized as the “3 P’s” model: Personalization, Predictability, and Proactivity. With this, an improved version of ChatGPT or a similar technology, will be tailored to the individual and their needs, sharing knowledge or taking action with the user’s permission, without requiring an explicit query.

Last point, assessing ChatGPT’s blind spots and understanding how it could be used for harmful purposes is probably one of the reasons OpenAI is offering it for free on the Internet. The next version will certainly be a paid version. Sam Altman, the start-up’s CEO, said that OpenAI will seek to monetise the model by charging a few cents per chat. He also added that monetisation is becoming particularly important as “the computational costs are exorbitant”. For now, the buzz created around ChatGPT not only allows OpenAI to have free testers who will help to improve the tool but will also help convince new investors to invest in this extremely promising project. 

The limits of the tool.

The most important reservation we can make about this tool surely lies in the quality of the content itself. In his December 11 tweet, Sam altman, Co-Founder & CEO of OpenAI, warned users about the data provided by the tool.

This information is also relayed on the homepage of the OpenAI website states that the chatbot can generate “incorrect information” or “produce dangerous instructions or biased content”.

In fact, ChatGPT elaborates its answers through texts that have a high probability of being recognized by a human as an answer. However, it is not uncommon for the A.I. to not have the answer to the question asked. When it is asked a question, it will build a text based on the contents available in its “learning data”. When the question is “simple”, in other words, if there are many texts that answer it in its learning database, there is a very high probability that its answer will be factually correct. As soon as you deviate a little from the easy cases, on the other hand, ChatGPT will fill in the gaps by inventing what it is missing.

Secondly, we can mention the problem of copyright. The system gets its inspiration in real texts to generate the content. It is then very difficult to cross-check the information to find the original in order to quote its author.

Then, ChatGPT refuses to take sides. When faced with a sensitive issue, it will approach it carefully and diplomatically. The developers have put in place numerous safeguards preventing the tool from answering questions related to racism, politics or other offensive outbursts. If it nevertheless speaks out on one of these sensitive issues, you will get an unbiased summary of the views of each party.

Finally, some more skeptical analysts question the deep learning mechanism that fuels the tool. According to them, it is a bit too early to assert that ChatGPT is an example of advanced deep learning, which suggests that the machine is capable of improving itself.

According to them, at this moment, ChatGPT does not have any particular capability in this regard. The tool is not able to touch its own code, or to do anything autonomously. All of its “improvements” at this point come from human feedback that tells it when its answers are wrong or when they are convincing. Some observers even claim that OpenAI’s goal is to generate buzz around the release of the tool so that all users provide masses of learning data for free that would cost them millions if they had to pay people to do it…

In short, we recommend that users use the tool as a source of creative inspiration, to give ideas and serve as a starting point for a work that will afterwards mainly be human. In other words, ChatGPT can give a lot of ideas that will be a mixture of existing things and that can perhaps take us in a direction that we would not have thought of alone.
Ultimately, ChatGPT is a tool that can help the users in their daily life, but it will never replace the added value that a human being can bring.

Practical examples for a sensible use of ChatGPT.

That said, not everything has to be thrown away though. The tool can be very useful for some tasks but the requests must respect some basic tips allowing to get the best of its capabilities.

The very first advice we can give is to add context to the query and be specific.  In this way the user enhance the chance to receive more relevant, customized, and valuable answers.
So, it’s recommended  to had parameters in the query : Ask a question, and at the end, add restrictions or requirements. For instance : “the text must count 200 words”, “The word architect must be included in the text”, “give me a step-by-step detail”, “keep the project under 500€”…

Then, since this is a conversational tool, it is possible to have the answer you get modified thanks to commenting the received answer.  Examples: make it funnier, longer, more technical, use fewer adjectives, add more hashtags,…

Further, preparing for the future means understanding it. It is thus recommended to try to break the tool and to observe the various flaws of it : bias, data privacy, accuracy, reliance, plagiarism,…

Finally, and good to know, it is possible to ask the tool to provide an answer under the form of a table, a list,a chart or even a code. 

Here are some examples of how to get the most out of the tool and how to make gains in productivity and creativity.

As a chatbot in customers servicing.


As explained earlier, ChatGPT can generate conversations from just a few words of input from a user. This makes it ideal for use in applications such as chatbots, virtual assistants. The tool is thus particularly useful in customer service applications, as it can quickly generate natural language responses to customer queries. This enables customers to get the help they need without having to wait for a human agent to respond.
In a similar vein, it’s possible to let tool answer social media comments and reviews.

Coding and debugging.


Another skill that will certainly be highly appreciated by software developers is the tool’s ability to write lines of code. ChatGPT can write a code snippet if the users describe the problem statement correctly. But that’s not all, ChatGPT can help you deduce bugs in your code with just one click. It will not only highlight the error, but also present you with a detailed solution to fix it.

Summary and paraphrasing.


It is totally possible to copy paste a text and to let ChatGPT summarize it or paraphrase it. This can be very useful when receiving a long report to summarize the notes from a meeting for instance.
Plus, ChatGPT’s bot is able to explain technical concepts (related to mathematics, science or engineering for instance) and to translate them into simple and intelligible terms for everyone.

Enhancing creativity and get ideas


It is sometimes extremely difficult to find what to make for dinner, what to give as a gift for Christmas or what content to publish on social networks. ChatGPT offers various and concrete answers that save us a lot of time. The same goes for writing an email when we lack inspiration or, more creatively, writing a story for children. Many examples exist on the internet where users generate complete stories based on a few keywords provided to the tool.
The same goes for sports sessions. The AI is able to develop detailed and super comprehensive programs.

For Marketers


Another interesting opportunity offered by the tool is the possibility to generate a list of keywords to be well referenced on search engines for example and enhance a SEO strategy. Further, it can be useful to let ChatGPT generate meta titles and meta descriptions based on a given content.

Also very useful to generate the structure of an article. The tool is able to prioritize the ideas and sort out what is most relevant and what is less relevant. Thus, it offers us H1, H2, H3, ….
For a product page, it can write a sales pitch or a complete product description. Plus, why not let the FAQs be written ?

For Project Managers


ChatGPT can Generate the first draft of a schedule.  After a meeting, the tool can summarize the notes and create action items when necessary. Finally, it can help for the prioritization of tasks to balance multiple stakeholders and their needs.
We could for instance ask: “Create a schedule in order to launch my new startup proposing carbon neutral personalized goodies for companies by December 1st, 2023. Please include deliverables, timelines, contingency planning, team bonding, breaks, brainstorming, and user testing.”

To conclude :

In closing, here is a final line of thought. Artificial intelligence is fine, but what kind of intelligence are we talking about here?

It is legitimate to be enthusiastic about this new technologies, but it is important to take a step back from it and to question how it works in order to get the best out of it and keep a critical eye.

GPT is an intelligent machine, but in the sense of an intelligence that mobilizes knowledge, itself reduced to language sets. The intelligence here is therefore an intelligence that analyzes signs, words and combines them based on statistical regularities.

Nevertheless, as we all know, this intelligence can not be reduced to this art of mathematical combination. Intelligences are multiple : There is an intelligence of sensitivity, an intelligence of intuition, a moral intelligence,… We mobilize these in turn when we make judgments, when we express intentions, convictions, feelings… All these judgments, GPT is not able to write them down.  It is therefore important to understand that our intelligence is not only a matter of calculations, but also of subjective singularity. This allows us to question ourselves, to be critical, to grasp the particular reasoning of an emotion.

We have thus entered a new area, that of a dialogue between an artificial intelligence on the one hand and on the other hand, ourselves, multiple and lived intelligences. For the moment, AI’s remain tools that can be used to increase productivity and even creativity but will never replace the work of the human being endowed with sensitivity and multiple intelligences.

Critical thinking is thus vital with the rise of generative AI. The best advice we can give is to try out and explore these technologies, and to stay conscious of their limitations. Tread carefully when it comes to bias, data privacy, and the objectives of the creators.

Endnote:

The images that illustrate this article were also generated using artificial intelligence tools. You can also create your own thanks to platforms such as DALL.E 2, another platform developed by OpenAI.


Sources:

Veuillez saisir votre adresse email pour vous abonner. Envoyer