At the end of November, research organization OpenAI released a first beta version of ChatGPT, a new chatbot based on artificial intelligence. To conduct conversations with humans, and as soon as it was tried, the new robot embarked on a journey through social networking sites, especially on the Twitter platform, where Sam Altman, CEO of the organization, stated that one million users tried the new robot within one week of its launch.
ChatGPT launched on wednesday.
Today it has crossed 1 million users!
- Sam Altman (@sama) December 5, 2022
OpenAI is the same organization that earlier launched Dall-E, which turns users' words into images and creative artwork using artificial intelligence, and it emerged in 2015 with funding from some of Silicon Valley's most prominent billionaires. The most famous of them is Elon Musk, who resigned three years later from the company's management, with the aim of establishing a non-profit organization focused on developing artificial intelligence.
(1)
Over the past days, hundreds of screenshots have spread of discussions with the "Chat GBT" chatbot on the Twitter platform as he talks about almost all topics, and many of the first users were amazed at what his ability to discuss, answer questions and write has reached, so that he He wrote verses of poetry, scholarly articles, and some jokes, so much so that some considered him to be a mixture of software and magic!
Chat with a robot!
(Shutterstock)
Over the past decade, most chatbot experiments have often failed, and it's not impressive unless you pick the best responses for the bot and discard the rest.
In recent years, some artificial intelligence software has emerged that has succeeded in specific tasks, such as writing marketing content, but is mostly limited to this task only, and cannot add anything new outside the scope specified for it.
But ChatGBT looks different, perhaps smarter, weirder, and more agile than previous bots.
The goal of the GBT Chatbot project is to try to make speech using artificial intelligence sound more fluid and natural, just as human conversations do.
Users can ask the bot their questions, and it will answer them with complete sentences, trying to mimic the cadence of a natural conversation with a real person.
But the organization constantly warns that the answers are not always correct or appropriate.
(2)
The organization's philosophy is to release these prototypes to the general public, and before putting in place all the firewalls, in the hope that user feedback, feedback, and ratings will help us find and address issues based on the bot's interaction in the real world, which seems to be happening.
While many of the screenshots of the conversations with the bot that spread on Twitter were strange and exciting conversations, users also found useful applications for these conversations, as some believe that it is the first chat bot that is fun enough to talk to, and useful enough to ask it for information, as well as He can participate in philosophical discussions and help in practical matters. For example, it seems that the robot can help programmers discover and fix errors in their code, and many have asked him to write code in languages such as Python.
ChatGPT could be a good debugging companion;
it not only explains the bug but fixes it and explains the fix 🤯 pic.twitter.com/5x9n66pVqj
— Amjad Masad ⠕ (@amasad) November 30, 2022
The robot can also explain complex scientific terms or concepts in a simple way to a young child. It also seems to be good at answering open-ended analytical questions that often appear in homework. For this reason, users asked it to write articles suitable for study on specific topics, and the result was impressive. Some writers even expected academic articles to end in their current form due to artificial intelligence techniques.
(3)
I guess GPT-3 is old news, but playing with OpenAI's new chatbot is mindblowing.
https://t.co/so1TuXMQB0
We're witnessing the death of the college essay in real time.
Here's the response to a prompt from one of my 200-level history classes at Amherst
Solid A- work in 10 seconds pic.twitter.com/z1KPxiAc1O
— Corry Wang (@corry_wang) December 1, 2022
Learning from humans!
The technology that supports the chatbot is not entirely new, as it is based on the "GPT-3.5" text generator, the modern version of the "GPT-3" version, which also raised a wave of surprise when it was introduced by "Open AI" in 2020. (4) The organization is among the many companies, academic labs, and independent researchers who have been working for years to create more advanced chatbots.These systems certainly can't talk quite like humans, but they often look like they do, or maybe that's their future goal, some even speculate. It will replace the traditional search engines that we know like Google and others.
Google itself has recently been working on its own conversational system, which is known as "LaMDA", and this system sparked great controversy last June after one of the company's engineers claimed that the artificial intelligence he was talking to had human consciousness, of course it was not. Also, but it captured the public's imagination even though Google did not release this system for the public experience.
(5)
In 2018, researchers at Google and labs such as OpenAI began designing neural networks that analyze huge amounts of digital text, including books, Wikipedia articles, news and online chat logs, in what is known as "large language models". With billions of distinct patterns in the way people associate words, numbers, and symbols, these systems learn to compose texts and generate responses on their own.
(6)
In this sense, chatbots are essentially "large language models," and they use algorithms to analyze the huge body of text they collect from these online sources, to answer user requests in what can be human-sounding language.
However, most chatbots do not have memory and are not programmed to remember or learn from previous conversations, meaning that they treat each new request as a completely new page, and this is precisely what distinguishes the "GBT Chat" bot from others.
(Shutterstock)
The new robot remembers what the user said previously, because it relies on a machine learning technique known as "Reinforcement Learning from Human Feedback", a technique that allows software to learn through human feedback and evaluations.
(7) After the person has tested the robot, he is asked to rate his responses: Were they convincing, useful, or correct?
Then, using reinforcement learning technology, the robot takes advantage of those assessments to fine-tune its system and more carefully decide what it will and won't do.
In a simple way, we can describe the "GBT Chat" robot as an artificial intelligence system trained to recognize patterns from huge and huge segments of texts taken from the Internet, and then trained more with human assistance to provide better conversation and more useful dialogue with humans.
The answers you get may seem convincing and even reliable, but they may be completely wrong answers, which is what the "Open AI" organization warns about.
Blind spots for gpt chat!
This alerts us that the new chatbot is by no means perfect, as it appears at first glance, because the way it works and generates answers often makes it prone to providing wrong answers, even in mathematical problems that seem simple to humans.
On December 5, the administrators of Stack Overflow, one of the most famous sites that gather programmers in the world, prevented users from sharing the responses produced by the ChatGBT chatbot, and made it clear that the bot makes it very easy for users to Giving ready and quick answers and flooding the site with them, which may seem correct at first glance but are often wrong when examined carefully.
(8)
Unlike Google, the "GBT Chat" robot does not have the ability to crawl various websites to obtain information about current events, and its knowledge is limited to things it learned previously (before 2021), which makes some of its answers seem outdated, even Sam Altman He mentioned in a tweet on his Twitter account, warning that the capabilities of the robot are very limited, and it is wrong to rely on it for anything important at the present time.
ChatGPT is incredibly limited, but good enough at some things to create a misleading impression of greatness.
it's a mistake to rely on it for anything important right now.
it's a preview of progress;
we have lots of work to do on robustness and truthfulness.
- Sam Altman (@sama) December 11, 2022
To add a touch of safety to its use, the bot was programmed to reject "inappropriate requests" by default, such as giving answers to illegal activities, but users have found ways to bypass many of these barriers, including reformulating a request for answers to illegal activity as an experiment. a theoretical idea, or asking him to write a scene from a play that includes this activity, or even instructing the robot to disable its security features.
Assessment of these bot programming blind spots and how they can be misused for illegal or malicious purposes is presumably a key part of the goals of OpenAI to release the bot version to the general public for testing, and future releases will often close such vulnerabilities, as well as Other vulnerabilities you haven't discovered yet.
Artificial intelligence for the public!
While a highly efficient language bot may be familiar to AI researchers, it is almost the first time that such a powerful tool has been made available to the general public via a free, easy-to-use interface.
Because all you need is to create an account on the site and start using and experimenting with the new robot, but noting that the services of the "Open AI" organization have not reached all Arab countries yet.
Much the same dynamic is evident in the proliferation of AI image generators.
Again, these systems have been in development for years, but access has not been possible.
But this year, systems like Midjourney and Stable Diffusion allowed anyone to use that technology for free and easily, and suddenly AI art and paintings are everywhere on the internet.
View this post on Instagram
A post shared by Mohamad Yousef (@mohamad.a.yousef)
Almost simultaneously, LensAI, an AI photo editing app on iPhone and Android devices, has become everyone's new favorite artist lately.
This is why you will find that many people in their social media profile pictures suddenly look like anime characters or oil paintings.
The app also uses a technology called "Stable diffusion" from image-generating startup Stability AI, and turns the photos you give it into artwork.
(9)
These products not only tell us about the extent of the development of artificial intelligence technologies, but they also prove that the use of these technologies has become available to the general public and non-specialists, and that it is no longer confined to laboratories and large companies only, which can change our lives in the future, in a way that no one could have imagined. Imagine it years ago.
_________________________________
Sources:
(1) Introducing OpenAI
(2) ChatGPT: Optimizing Language Models for Dialogue
(3) The College Essay Is Dead
(4) Meet GPT-3.
It Has Learned to Code (and Blog and Argue)
(5) From the China Room to see bats to the world..Has Google developed a conscious robot like us?
(6) Finally, a Machine That Can Finish Your Sentence
(7) ChatGPT: Optimizing Language Models for Dialogue
(8) Temporary policy: ChatGPT is banned
(9) Stable Diffusion Public Release