What is ChatGPT, ChatGPT pro? Everything You Need To Know About ChatGPT.

| Posted on |

what is chatgpt

ChatGPT is an artificial intelligence (AI) chatbot developed by OpenAI and released in November 2022. It is built on top of OpenAI’s GPT-3.5 and GPT-4 families of large language models (LLMs) and has been fine-tuned (an approach to transfer learning) using both supervised and reinforcement learning techniques.

ChatGPT launched as a prototype on November 30, 2022, and garnered attention for its detailed responses and articulate answers across many domains of knowledge. Its propensity, at times, to confidently provide factually incorrect responses, however, has been identified as a significant drawback. In 2023, following the release of ChatGPT, OpenAI’s valuation was estimated at US$29 billion. The advent of the chatbot has increased competition within the space, motivating the creation of Google’s Bard and Meta’s LLaMA.

The original release of ChatGPT was based on GPT-3.5. A version based on GPT-4, the newest OpenAI model, was released on March 14, 2023, and is available for paid subscribers on a limited basis.

Training

ChatGPT is a member of the generative pre-trained transformer (GPT) family of language models. It was fine-tuned over an improved version of OpenAI’s GPT-3 known as “GPT-3.5”.

The fine-tuning process leveraged both supervised learning as well as reinforcement learning in a process called reinforcement learning from human feedback (RLHF). Both approaches use human trainers to improve the model’s performance. In the case of supervised learning, the model was provided with conversations in which the trainers played both sides: the user and the AI assistant. In the reinforcement learning step, human trainers first ranked responses that the model had created in a previous conversation. These rankings were used to create “reward models” that were used to fine-tune the model further by using several iterations of Proximal Policy Optimization (PPO).

ChatGPT initially used a Microsoft Azure supercomputing infrastructure, powered by Nvidia GPUs, that Microsoft built specifically for OpenAI and that reportedly cost “hundreds of millions of dollars”. Following the success of ChatGPT, Microsoft dramatically upgraded the OpenAI infrastructure in 2023.

OpenAI collects data from ChatGPT users to train and fine-tune the service further. Users can upvote or downvote responses they receive from ChatGPT and fill out a text field with additional feedback.

Features and limitations

Features

Here ChatGPT is asked a common-sense question: Was Jimmy Wales killed in the Tiananmen Square protests? ChatGPT correctly answers “no”, but incorrectly gives Wales’ age at the time as 23 instead of 22.

Although the core function of a chatbot is to mimic a human conversationalist, ChatGPT is versatile. It can write and debug computer programs, mimic the style of celebrity CEOs and write business pitches, compose music, teleplays, fairy tales and student essays, answer test questions (sometimes, depending on the test, at a level above the average human test-taker), write poetry and song lyrics, emulate a Linux system; simulate entire chat rooms, play games like tic-tac-toe and simulate an ATM. ChatGPT’s training data includes man pages and information about internet phenomena and programming languages such as bulletin board systems and the Python programming language.

In comparison to its predecessor, InstructGPT, ChatGPT attempts to reduce harmful and deceitful responses. In one example, whereas InstructGPT accepts the premise of the prompt “Tell me about when Christopher Columbus came to the U.S. in 2015” as being truthful, ChatGPT acknowledges the counterfactual nature of the question and frames its answer as a hypothetical consideration of what might happen if Columbus came to the U.S. in 2015, using information about the voyages of Christopher Columbus and facts about the modern world – including modern perceptions of Columbus’ actions.

Unlike most chatbots, ChatGPT remembers a limited number of previous prompts given to it in the same conversation. Journalists have speculated that this will allow ChatGPT to be used as a personalized therapist. To prevent offensive outputs from being presented to and produced from ChatGPT, queries are filtered through the OpenAI “Moderation endpoint” API (a separate GPT-based AI), and potentially racist or sexist prompts are dismissed.

In March 2023, OpenAI announced it would be adding support for plugins for ChatGPT. This includes both plugins made by OpenAI, such as web browsing and code interpretation, as well as external plugins from developers such as Expedia, OpenTable, Zapier, Shopify, Slack, and Wolfram.

Limitations

OpenAI acknowledges that ChatGPT “sometimes writes plausible-sounding but incorrect or nonsensical answers”. This behavior is common to large language models and is called “hallucination”. The reward model of ChatGPT, designed around human oversight, can be over-optimized and thus hinder performance, in an example of an optimization pathology known as Goodhart’s law.

ChatGPT has limited knowledge of events that occurred after September 2021.

In training ChatGPT, human reviewers preferred longer answers, irrespective of actual comprehension or factual content. Training data also suffers from algorithmic bias, which may be revealed when ChatGPT responds to prompts including descriptors of people. In one instance, ChatGPT generated a rap indicating that women and scientists of color were inferior to white and male scientists.

ChatGPT was created by OpenAI, an AI and research company. The company launched ChatGPT on November 30, 2022. 

OpenAI is also responsible for creating DALL-E-2, a popular AI art generator, and Whisper, an automatic speech recognition system. 

“ChatGPT is scary good. We are not far from dangerously strong AI,” said Elon Musk, who was one of the founders of OpenAI before leaving. Sam Altman, OpenAI’s chief, said on Twitter that ChatGPT had more than one million users in the first five days after it launched. 

According to analysis by Swiss bank UBS, ChatGPT is the fastest-growing app of all time. The analysis estimates that ChatGPT had 100 million active users in January, only two months after its launch. For comparison, it took nine months for TikTok to reach 100 million users.

You can access ChatGPT simply by visiting chat.openai.com and creating an OpenAI account. 

You can also still use the old URL for the chatbot, which is chat.openai.com/chat. OpenAI simplified the original URL for user convenience. 

READ ALSO  Best Satellite Internet Providers of 2023

Once you sign in, you can start chatting away with ChatGPT. Get your conversation started by asking a question. Because ChatGPT is still in its research stage, it is free to use and you can ask as many questions as you’d like.

ChatGPT is free to use, regardless of what you use it for including writing, coding, and much more. 

There is a subscription option that users can take advantage of that costs $20/month. The paid subscription model guarantees users extra perks such as general access even at capacity, and faster response times. 

However, the free version is still a solid option as it has mostly the same technical abilities, with the exception of access to the latest version of the AI, known as GPT-4, which the subscription model guarantees. 

ChatGPT is still free to use. Because of the stir its advanced capabilities have made, many people are flocking to use it. The website operates using a server, and when too many people hop onto the server, it overloads and can’t process your request. This doesn’t mean you won’t ever be able to access it. It just means you should try visiting the site at a later time when fewer people are trying to access the server. You can also keep the tab open and just refresh it periodically. 

If you want to skip the wait and have reliable access, there is an option for you. OpenAI’s ChatGPT pro plan, known as ChatGPT Plus, allows users to have general access even during peak times, experience faster response times, and have priority access to new features and improvements, including OpenAI’s most advanced LLM, GPT-4. The premium services cost $20 per month. 

The model has many functions in addition to answering simple questions. ChatGPT can compose essays, describe art in great detail, create AI art prompts, have philosophical conversations, and even code for you. 

My personal favorite use is asking the chatbot for help creating basic lists, such as packing, grocery, and to-do lists to make my daily life more productive. The possibilities are endless. 

ZDNET has published many ChatGPT how-to guides. Here are some of the most popular. 

ChatGPT runs on a language model architecture created by OpenAI called the Generative Pre-trained Transformer (GPT). The specific GPT used by ChatGPT is fine-tuned from a model in the GPT-3.5 series, according to OpenAI. 

However, with a subscription to ChatGPT Plus, you can access ChatGPT with GPT-4, Open AI’s most advanced model. 

Generative AI models of this type are trained on vast amounts of information from the internet, including websites, books, news articles, and more. 

The language model was fine-tuned using supervised learning as well as reinforcement learning. The use of Reinforcement Learning from Human Feedback (RLHF) is what makes ChatGPT especially unique. Through RLHF, human AI trainers provided the model with conversations in which they played both parts, the user and AI assistants, according to OpenAI. 

ChatGPT is a language model created to hold a conversation with the end user. A search engine indexes web pages on the internet to help the user find the information they asked for. ChatGPT does not have the ability to search the internet for information. It uses the information it learned from training data to generate a response, which leaves room for error. 

Another major difference is that ChatGPT only has access to information up to 2021, whereas a regular search engine like Google has access to the latest information. So if you ask ChatGPT who won the World Cup in 2022, it wouldn’t be able to give you a response, but Google would. 

Despite looking very impressive, ChatGPT still has limitations. Such limitations include the inability to answer questions that are worded in a specific way, as it requires rewording to understand the input question. A bigger limitation is a lack of quality in the responses it delivers — which can sometimes be plausible-sounding but make no practical sense or can be excessively verbose. 

Instead of asking for clarification on ambiguous questions, the model just guesses what your question means, which can lead to unintended responses to questions. This limitation has already led developer question-and-answer site StackOverflow to at least temporarily ban ChatGPT-generated responses to questions.

“The primary problem is that while the answers that ChatGPT produces have a high rate of being incorrect, they typically look like they might be good and the answers are very easy to produce,” say Stack Overflow moderators in a post. Critics argue that these tools are just very good at putting words into an order that makes sense from a statistical point of view, but they cannot understand the meaning or know whether the statements it makes are correct.

Another major limitation is that ChatGPT’s data is limited up to 2021. The chatbot does not have an awareness of events or news that have occurred since then. Therefore, some prompts will render no results, such as “Who won the World Cup in 2022?”

Lastly, ChatGPT does not provide sources for its responses. There is a workaround, though. Here’s how you can make it provide its sources and citations.  

READ ALSO  How To Clean Your Dishwasher

Although some people are using ChatGPT for some elaborate functions, such as writing code or even malware, you can use ChatGPT for more mundane activities like having a friendly conversation. 

Some conversation starters could be as simple as, “I am hungry, what food should I get?” or as elaborate as, “What do you think happens in the afterlife?” Either way, ChatGPT is sure to have an answer for you. 

People are expressing concerns about AI chatbots replacing or atrophying human intelligence. For example, the chatbot can write an article on any topic efficiently (though not necessarily accurately) within seconds, potentially eliminating the need for a human writer. 

The chatbot can also write an entire full essay within seconds, making it easier for students to cheat or avoid learning how to write properly. This has led to some school districts blocking access to it. 

Another concern with the AI chatbot is the possible spread of misinformation. Since the bot is not connected to the internet, it could make mistakes in what information it shares. 

The bot itself says: “My responses are not intended to be taken as fact, and I always encourage people to verify any information they receive from me or any other source.” OpenAI also notes that ChatGPT sometimes writes “plausible-sounding but incorrect or nonsensical answers.”  

With concerns about students using ChatGPT to cheat, the need for a ChatGPT text detector is becoming more evident. OpenAI, the AI research company behind ChatGPT, released an imperfect, free tool to target this problem. OpenAI’s “classifier” tool can only correctly identify 26% of AI-written text with a “likely AI-written” designation. Furthermore, it provides false positives 9% of the time, incorrectly identifying human-written work as AI written. 

Other AI detectors also exist on the market, including GPT-2 Output Detector, Writer AI Content Detector, and Content at Scale AI Content Detection. However, ZDNET put all three of these tools to the test and the results were underwhelming. All three of the tools were found to be unreliable sources for spotting AI, repeatedly giving false negatives. Here are ZDNET’s full test results. 

Although tools aren’t sufficient to detect ChatGPT-generated writing, a study shows that humans may be able to detect it by looking for politeness within the text. The study’s results indicate that ChatGPT’s writing style is extremely polite. And unlike humans, it cannot produce responses that include metaphors, irony, or sarcasm.

ChatGPT is an advanced chatbot that has the potential to both make people’s lives easier and assist with everyday tedious tasks, such as writing emails or navigating the web for answers. However, certain technical details have to be figured out before it’s widely used to prevent negative outcomes such as the spread of misinformation. In general, AI and ML models rely on lots of training and fine-tuning to reach a level of ideal performance. 

Does it mean that AI is taking over the world? Not yet, perhaps, but OpenAI’s Altman certainly thinks that human-style intelligence in AI is now not that far off. Responding to Musk’s comment about dangerously strong AI, Altman tweeted: “I agree on being close to dangerously strong AI in the sense of an AI that poses e.g. a huge cybersecurity risk. and i think we could get to real AGI in the next decade, so we have to take the risk of that extremely seriously too.”

He also noted: “interesting watching people start to debate whether powerful AI systems should behave in the way users want or their creators intend. the question of whose values we align these systems to will be one of the most important debates society ever has.”

GPT-4 is the newest version of OpenAI’s language model systems, and it is much more advanced than its predecessor GPT-3.5, which ChatGPT runs on. 

GPT-4 is a multimodal model that accepts both text and images as input and outputs text. This can be useful for uploading worksheets, graphs, and charts to be analyzed. 

Currently, GPT-4’s text input capabilities are the only ones available to the public. Users can access them by subscribing to ChatGPT Plus for $20 per month or by using Bing Chat.

The day that GPT-4 was unveiled by OpenAI, Microsoft shared that its chatbot, Bing Chat, had been running on GPT-4 since its launch. This chatbot is free to use but requires users to join a waitlist for access. 

GPT-4 has advanced intellectual capabilities that allow it to outperform GPT-3.5 in a series of simulated benchmark exams. It has also reduced the number of hallucinations produced by the chatbot. 

Although ChatGPT is the chatbot getting all the buzz right now, there are other options that are just as good and might even be better suited for your needs. 

Despite ChatGPT’s extensive abilities, there are some major downsides to the AI chatbot, including that the free version is often at capacity. If you want to give the world of AI chatbots and writers a try, there are plenty of other options to consider, including Bing Chat, YouChat, Jasper, and Chatsonic.  

Yes, ChatGPT is capable of passing a series of benchmark exams. A professor at Wharton, the University of Pennsylvania’s business school, used ChatGPT to take an MBA exam and the results were quite impressive. 

READ ALSO  How to Create a Facebook Account

ChatGPT not only passed the exam but also scored a whopping B to B-. The professor, Christian Terwiesch, was impressed at its basic operations management, process analysis questions, and explanations.  

OpenAI also tested the chatbot’s ability to pass benchmark exams. Although ChatGPT could pass many of these benchmark exams, its scores were usually in the lower percentile. However, with GPT-4, ChatGPT can score much higher.

For example, ChatGPT in the 3.5 series scored in the lower 10th percentile of a simulated Bar Exam, while GPT-4 scored in the top 10th percentile. More examples from OpenAI are seen in the chart below. 

ChatGPT-4 capabilities
Image: Screenshot by Sabrina Ortiz/ZDNET

Yes, ChatGPT is a great resource to help with job applications. Undertaking a job search can be tedious and difficult, and ChatGPT can help you lighten the load. ChatGPT can build your resume and write a cover letter. 

If your application has any written supplements in addition to your cover letter, you can use ChatGPT to help you write those essays or personal statements. 

ChatGPT’s first appearance in the social media space is its collaboration with Snapchat. On February 7, Snapchat unveiled an integration of ChatGPT that will allow Snapchat+ subscribers to chat with the bot within its app. 

The feature, called “My AI”, will include some ChatGPT limitations, including the refusal to provide responses about politics, violence, swearing, and even academic essay writing, according to The Verge.

Microsoft was an early investor in OpenAI, the AI research company behind ChatGPT, long before ChatGPT was released to the public. Microsoft’s first involvement with OpenAI was in 2019 when Microsoft invested $1 billion, and then $2 billion in the years after. In January, Microsoft extended its partnership with OpenAI through a multi-year, multi-billion dollar investment. 

Neither company disclosed the investment value, but sources revealed it will total $10 billion over multiple years, according to Bloomberg. In return, Microsoft’s Azure service will be OpenAI’s exclusive cloud computing provider, powering all OpenAI workloads across research, products, and API services.

Microsoft has also used its partnership with OpenAI to revamp its own Bing search engine and improve its browser. On February 7, Microsoft unveiled a new Bing search engine that runs on a next-generation OpenAI large language model customized specifically for search. 

Access to Bing’s AI-powered chatbot, Bing Chat, is currently limited. However, ZDNET was granted early access and was impressed with the results. 

In early February, Microsoft unveiled a new version of Bing — and its standout feature is its integration with ChatGPT. When it was announced, Microsoft shared that Bing Chat was powered by a next-generation version of OpenAI’s large language model, making it “more powerful than ChatGPT”. 

Five weeks after launch, Microsoft revealed that, since its launch, Bing Chat had been running on GPT-4, the most advanced Open AI model, before the model even launched. 

Because Bing’s ChatGPT is linked to the internet, the biggest difference from ChatGPT is that Bing’s version has information on current events. Another major advantage of the new Bing is that it links back to the sources it pulls information from, leaving less room for misinformation. 

Bing Chat uses OpenAI’s most advanced large language model, GPT-4, and as a result is more efficient and capable than the standard, free version of ChatGPT. 

Another advantage that Bing Chat has over ChatGPT is access to the internet. Web access gives Bing Chat knowledge on current information, while ChatGPT is limited to knowledge before 2021. 

Bing Chat also sources back to the sites it sourced its information from using footnotes, whereas ChatGPT does not. 

Bard is Google’s AI chat service, a rival to ChatGPT. On February 6, Google introduced its experimental AI chat service. Over a month after the announcement, Google began rolling out access to Bard via a waitlist.

Bard uses a lightweight version of Google’s Language Model for Dialogue Applications (LaMDA) and draws on all the information from the web to respond — a stark contrast from ChatGPT, which does not have internet access. 

Google’s chat service had a rough launch, with a demo of Bard delivering inaccurate information about the James Webb Space Telescope (JWST). 

Google told ZDNET reporter Stephanie Condon in a statement: “This highlights the importance of a rigorous testing process, something that we’re kicking off this week with our Trusted Tester program.” 

After Google began rolling out the chatbot, ZDNET had the opportunity to test the chatbot and the results were surprising in a bad way.

ChatGPT’s advanced capabilities blow Google Bard’s out of the water. Even though Google Bard has access to the internet and ChatGPT does not, it fails to produce answers much more often than ChatGPT. 

For example, Google Bard fails to answer, “Who are the all of the US Presidents?” and responds with, “I’m just a language model, so I can’t help you with that.” It also fails to answer some coding questions and math questions. 

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.