ChatGPT: An Explainer

The internet is enchanted with ChatGPT (Chat Generative Pre-trained Transformer) at the moment — an AI chatbot developed by the OpenAI foundation and launched as a prototype in November 2022.

In two months, ChatGPT gathered over 100 million users, making it the fastest-growing consumer application — for its ability to provide fluent and comprehensive answers to queries on a wide range of subject fields in a human-like conversational manner — in contrast to how modern search engines like Google responds to internet searches with a list of blue-coloured links.

Currently available for anyone to use for free via a web page, the AI chatbot is trained to cater to the queries asked based on a huge sample of internet text.

ChatGPT’s current knowledge is limited to world events until 2021, yet it is growing in popularity and usage. As such, users now frequently encounter a notice informing them that ChatGPT is full and unable to process requests. 

Tech behind ChatGPT
Put simply, the tech behind ChatGPT is AI — a simulation of human intelligence by machines, which functions by processing enormous internet data sets, analyzing them for patterns, and using identified patterns to anticipate potential future states.

Some specific applications of AI include Expert systems — a program that uses databases of expert knowledge to offer human expert-like services, for instance in the field of medical diagnosis; Natural Language Processing (NLP) that uses natural language data to process human language (such as Siri and Alexa); Speech recognition that converts spoken language into text forms; and machine vision that equips industrial equipment with the ability to see and analyze tasks in smart manufacturing, quality control, and worker safety.

In the case of ChatGPT, it is fine-tuned from the GPT-3.5 series Large Language Model (LLM) and Reinforcement Learning from Human Feedback (RLHF).

Training on LLM allows the program to precisely predict the next word in a sentence while RLHF can learn from human feedback and generate responses that are satisfactory to humans.

AI’s preceding ability to grasp human language was limited. However, the introduction of Large Language Models (LLMs) completely shifted the tide. LLMs are language models that use deep learning techniques to process large amounts of texts and understand their structures, meanings, and relationships between the words, predict and produce text or even images (as in the case of DALL-E-2). These models are believed to get smarter when trained with more datasets.

ChatGPT is based on GPT-3 LLM but it is not the only LLM that exists. A few others include ERNIE (Baidu), BLOOM (Huggingface), YaLM 100B (Yandex), Gopher (Deepmind/Google), and Megatron (NVIDIA and Microsoft) each developed under different research collaborations and using different datasets.

Earlier, OpenAI accessed Twitter’s database for training which Musk later stalled.

During the initial training, human AI trainers provided conversations in which they played both sides — the user and an AI assistant. “The trainers were given access to model-written suggestions to help them compose their responses. This new dialogue dataset was mixed with the previous dataset (InstructGPT) and was transformed into a dialogue format.”

There are now speculations that OpenAI is well into developing an even more advanced text-based AI program equipped with more data points — GPT4 — supposedly arriving next year.

Money talks
The developer of ChatGPT Open AI is a US non-profit AI research firm, established with a $1 billion investment pledge from individuals like Elon Musk, LinkedIn co-founder Reid Hoffman, and YC Research. Three years later, it turned into a ‘capped-profit’ lab (a hybrid of for-profit and non-profit) overseen by a nonprofit board citing reasons as “the need to invest billions of dollars in upcoming years into large-scale cloud computing, attracting and retaining talented people, and building AI supercomputers.”

In 2019, Microsoft backed OpenAI with a $1 billion investment and another round of investment in 2021, and now will be investing an additional $10 billion.

Although the base-level ChatGPT is now free for use, OpenAI’s CEO Sam Altman tweeted in December 2022 “We will have to monetize it somehow at some point; the compute costs are eye-watering.”

And in Feb 2023, the company expects to make $200 million in 2023, and $1 billion in 2024 piloted a subscription model launching ChatGPT Plus, starting at $20 per month.

How will generative AI change the world?

The development of a generative AI tool as ChatGPT is a display of how powerful modern AI has become and indicates possible disruption of how we retrieve information — including the launch of DALL-E-2, an AI image generator that functions similarly like ChatGPT generating artwork by typing instructions into a text box, also developed by Open AI.

Some see ChatGPT as an alternative to Google because it is capable of providing descriptions, answers, and solutions to complex questions including ways to write code and solve layout problems and optimization queries.

Together with other AI tools which are individually trained on large content data sets — text, images, audio, video — and collectively trending as GenAI, the generative AI tools have opened windows to wider applications in the fields such as media, sales, marketing, and advertising, gaming, law, healthcare and education, product design, graphic design, coding, architecture and more.

ChatGPT itself is anticipated to have applications in numerous fields such as customer support, marketing and sales, healthcare, education, creative works, and programming. Real-world use could include generating website content, writing essays and proposals, and providing responses and recommendations to customer inquiries. 

For instance, its possible application in healthcare can be in streamlining administrative activities like scheduling appointments and as far as offering a more accurate diagnosis based on vast medical records.

An editor in The Conversation phrases ChatGPT as “A high-level personal assistant to speed up bureaucratic tasks and increase time for patient interaction.”

Limitations
As with any new technology, it is imperative its limitations and potential for misuse are thoroughly considered.

Unlike humans, ChatGPT will lack the ability to fully comprehend the nuanced nature and meaning of words, human language, and conversation. Humans can understand, judge and interpret a question implicitly whereas ChatGPT takes it literally. 

Moreover, these chatbots have a way of molding what they have learned into something new, regardless of whether it is true, much like a good storyteller, observes a coverage in The New York Times. OpenAI itself admits that ChatGPT can give entirely wrong answers and present misinformation as fact, writing “plausible-sounding but incorrect or nonsensical answers.”

One prime concern, as different critics raise, is education and learning. 

The American philosopher and linguist Noam Chomsky described ChatGPT, as “basically high-tech plagiarism” and “a way of avoiding learning,” further adding that students using advanced technology to parry learning is “a sign that the educational system is failing.”

Professor Alex Lawrence at Weber State University called it “the greatest cheating tool ever invented.”

For this, educators are already finding different ways around this and banning the use of ChatGPT in schools.

New York City Schools, the largest school department in the US, have already forbidden its teachers and students from using the chatbot out of concern that the advanced AI will unleash a wave of cheating, which was then followed by Seattle Public Schools.

Dr. Andrew Rogoyski at the University of Surrey has stated that everything ChatGPT postulates is essentially a rehash of something that has been said before, by a human as it is a system that has memorized a billion books so that it can guess what comes after a question is asked.

In an effort to assuage some of the concerns regarding how people are exploiting text generation, ChatGPT launched its own plagiarism detection tool called 'AI text classifier.' However, the tool is still a work in progress and is imperfect, says OpenAI.

The bottom line is educators believe ChatGPT will be an impediment to developing critical thinking and problem-solving abilities.

On the other hand, an editorial at The Guardian questioned if any ChatGPT content ‘can be truly trusted’ and called for oversight since “giving out tools could be used to mass produce fake news and ‘trolling and griefing’ messages.”

Linguists have also warned that widespread use of ChatGPT could reduce linguistic as well as stylistic diversity and richness and reinforce the global dominance of English and some other European languages.

The race for AI chatbot
Despite the criticism, big tech firms are not shying away from developing their own AI chatbot.

Google launched an experimental service called 'Bard' in February 2023 after expressing concern over ChatGPT’s unexpected capability to threaten the search engine industry.

Given that Bard has access to the most recent information while ChatGPT is capped to 2021, Google clearly has an advantage in this context. For the time being, Google has opened up Bard to ‘trusted testers, and more and more users will have access to Bard for testing during the coming days. 

Sundar Pichai, CEO of Google, twitted, “As people turn to Google for deeper insights and understanding, AI can help us get to the heart of what they're looking for. Bard seeks to combine the breadth of the world's knowledge with the power, intelligence, and creativity of our large language models. It draws on information from the web to provide fresh, high-quality responses.”

But an error during the demo, where Bard gave inaccurate information about James Webb Space Telescope in response to a query, led Alphabet, Google’s parent company to lose $163 billion in market value.

The Chinese search engine firm, Baidu Inc. also announced in February 2023 that they too would launch a ChatGPT-style service soon. It shall be named "Wenxin Yiyan" in Chinese and ‘ERNIE Bot’ (Enhanced Representation through Knowledge Integration) in English.

The New York Times reported that other Chinese conglomerates like Alibaba, JD.com, and Tencent are also developing their own ChatGPT-type products, though details are scant and they have not provided exact implementation dates.

Hugging Face, an AI firm building a ChatGPT competitor, and Amazon, which has long been working on related technology, have announced cooperation.

SK Telecom of South Korea is working to launch “A.” – pronounced “A dot” – a super app that is based on generative AI, the technology behind ChatGPT.

The South Korean largest search engine firm, Naver has also announced that it would be launching a ChatGPT-style service called “SearchGPT” that specializes in queries in Korean. In the initial phase, instead of being included in Naver's search engine, SearchGPT will function as a separate beta service.

Meanwhile, the frenzy over generative AI has also heated up.

According to Nikkei Asia, the combined corporate value of generative AI companies has jumped sixfold in two years to $48 billion whereas OpenAI alone is valued at an estimated $29 billion. “Together, the startups are raising large sums of money to develop tools that can generate prose, imagery, or other content on command.”

For instance, Israeli startup AI21 Labs which launched an AI-based writing tool in January, citing its mission as providing new power to writers raised $64 million last year to develop a tool that can be used by a wide range of writers to edit and rewrite work.