Image source: pexels
ChatGPT is the new trending tech in the online landscape. But is it a fad or far from it? Academics, businesses, content creators, everyone has used the chatbot to get things done. But, are these tools worth the hype? We will learn that and more in this blog.
What is ChatGPT?
ChatGPT (short for Chat Generative Pre-Trained Transformer) is an AI-powered chatbot that is built to respond to users’ questions. The AI bot ChatGPT is an extremely efficient and powerful tool for getting answers on a wide range of topics.
Built in the tech hub of North California by OpenAI, ChatGPT was announced publicly in 2022. It was showcased as a language model trained on large text-based datasets from the internet. The AI chatbot caught enough fire online to get a legion of financial backers from diverse backgrounds.
Corporations like Microsoft invested in two rounds, first in 2019 with a 1 billion USD in 2019 to build ChatGPT’s artificial general intelligence (AGI); and then in January 2023, through another round of 10 billion USD approximately. The economics of which are incomprehensible but you be the judge.
What can ChatGPT do?
ChatGPT is really good at handling language and informational-oriented tasks. There are some TOS guidelines built into the system to prevent it from becoming a colossal disaster like Microsoft’s Tay AI in 2016, and ignore inappropriate requests.
After it became publicly accessible in 2022, the general populace of the internet used it for many things. As Statista informs, it took a mere 5 days for ChatGPT to hit the million-user mark. The hype did not fade, content creators across different fields used ChatGPT absolutely everything. Business and academic institutes tested its accuracy. Some of the notable exams it qualified include Google L3 coding, USMLE, Bar exam, AP English essay, etc.
However, it also failed a few, for example, standardized tests for IIT, Feinstein Institutes’ Gastroenterology exam, and UPSC of Indian Administrative Services.
However, keeping business aside, ChatGPT is a really fun tool to mess around and play with! Users can:
- Translate and learn languages. ChatGPT has an in-built language translator which works much better than Google translator.
- Play games. The AI ChatGPT is capable of playing a multitude of language-based games. Including Chess! It is a little bit tricky as you will have to use chess’ proper algebraic notation for proper coordination.
- Coding. The AI chatbot can help you create video games and even check bugs in your codes.
- Writing. ChatGPT allows people to generate a wide range of textual material such as poems, prose, articles, etc. The way ChatGPT is built helps it to go through massive amounts of text-based material to produce suggestions.
It can also write formal content such as resumes, cover letters, social media posts, etc.
How was ChatGPT trained?
ChatGPT training was done on large datasets of texts from the internet. According to Scott Aaronson, an OpenAI researcher, “(ChatGPT) simply consists of playing the following game over and over, trillions of times: predict which word comes next in this text string. So in some sense that’s its only goal or intention in the world: to predict the next word.”
Businesses that see ChatGPT as a trustworthy way to mechanize their workflow are in for an awakening. The most important thing to consider here is that ChatGPT is not a human but a bot placed under rigorous machine learning for a wide range of textual datasets. This can lead to complications in handling delicate issues, such as managing dialogue around marginalized communities. Here are some common risks that are involved with reliance on ChatGPT for business purposes-
Risk 1: Inaccurate and Insufficient Information Repository
To start it off, ChatGPT has a massive problem and it is that it has no new information after 2021. It does not have access to any new information, much less real-time data. The information available is lacking any new updates which puts a hard cap on the quality of responses you will get for your prompt. Unless you are searching for something very basic, you cannot trust it fully to deploy for your organizational matters. As it may skew results towards things you don’t want to.
Risk 2: ChatGPT lacks human nuance and understanding
The pre-training method of ChatGPT puts an instant limit on the kind of responses it can deliver. There are two gaping issues that the AI-powered chatbot faces:
- It does not have an “I don’t know” feature.
- The chatbot is not sensitive towards the nuances of language.
ChatGPT does not have an “I don’t know” response built into it which can result in the chatbot giving answers on topics which it has very little training. Such a level of self-awareness is a product of human understanding.
Another linguistic challenge that ChatGPT faces is that it does not understand sarcasm. Even simple sarcasm in sentences goes undetected by the system. An important observation made by researchers from the University of Ottawa was that generative text models such as ChatGPT struggle with idioms.
An idiomatic phrase such as, “Bite the bullet” or “break the bank” have figurative meaning(s) attached to them. Generally, the absence of idioms in a piece of written content signals that the content is AI-generated. ChatGPT is not human and certainly cannot converse in the human realm of interaction. The way it is trained and the way it operates severely restricts its ability to present thought in a unique way.
Risk 3: Contextual Understanding is a Limited
ChatGPT lacks the human nuance and contextual understanding of any given topic. Humans developing through knowledge is an understatement. Subjective viewpoints and individual experience are a major part of what constitutes humanity. It helps in developing complex and intriguing contextual comprehension of various social movements across histories and the development of philosophies.
ChatGPT summarizes a topic very well, but it fails to offer any singular or uncommon insights on the said topic. According to Professor Bartel, a professor of philosophy at Appalachian State University states, “They are really fluffy. There’s no context, there’s no depth or insight.”
ChatGPT limitations are that it cannot have a contextual understanding of a topic to generate original conclusions.
Risk 4: Little Expertise in the Domain
ChatGPT’s pre-training method makes it less prone to have in-depth information on any given topic. The academic community has been studying the viability of AI models like ChatGPT for assessment of research papers and it has routinely failed. (at the time of writing).
Training AI models on large datasets with very little influence on domain-specific expertise limits its ability to scrutinize ideas. It leads to ChatGPT struggling with providing businesses with anything that is going to materially change their operations.
Risk 5: Possibility of Biases & Inaccuracies
So far we have understood how the pre-training modeling limits ChatGPT’s capacity to provide contextual understanding but there is another problem with that 一 Bias. ChatGPT’s training on a fixed dataset puts a limit on the ideas it can conjure and gets compartmentalized sample bias. It does not help that ChatGPT uses predictive modeling to answer the prompts, where it uses the best word possible to create an answer.
This kind of methodology leaves a lot of space for inaccuracies. As it was reported by the Washington Post that ChatGPT is prone to hallucinations, where the AI-powered chatbot just makes random things up.
Risk 6: Demand for Human Supervision
ChatGPT requires constant human oversight to get all the results delivered by it in tune with human experience. All the answers have to be checked by a human being for
- Adding context
- Addressing biases
Human supervision ensures that ChatGPT is functioning and meeting the business requirements regularly and that is an additional work. ChatGPT as fickle and shallow as it is makes it an unreliable tool for businesses at large.
Risk 7: Data Leakages in ChatGPT
Recently ChatGPT has gotten into a cybersecurity controversy where bad actors used the AI-powered chatbot to write malicious code which allowed them to access users’ personal details. Important information such as users’ real names, email addresses , payment addresses , payment details, etc. were taken by the malignant actors.
The attack happened because of the open-source nature of the ChatGPT. It is not surprising that there has been a massive spike in security attacks on open-source libraries since 2019. However, OpenAI identified the issue immediately and fixed it within a few days, as per their statement “no important data was stolen by the miscreants.” But the questions around safety still remain answered as ChatGPT will have to constantly build an infrastructure to resist new forms of attack
Final Thoughts on ChatGPT
ChatGPT, as fun and informative as it seems, has serious shortcomings. It comes from two important factors:
- Its pre-training methodology where data is just fed into a system, and
- Lack of an artificial general intelligence (AGI) system.
The system is riddled with biases in the treatment of touchy subjects; the lack of contextual comprehension and domain expertise only makes it trickier to use as a daily business application. To make matters worse, the AI chatbot is not connected to the internet and does not have a database beyond 2021. This means any information on current events or developing subjects will not be available instantly.
Another major concern is server issues which cause ChatGPT down. ChatGPT network error on long responses happens routinely. However, to handle the ChatGPT down issue, OpenAI has started a paid service that costs $20 per month called ChatGPT Plus. An added benefit of ChatGPT Plus is that it is slightly more updated than the free version in terms of information.
ChatGPT as it stands now has a long way to become what its stated goals are. ChatGPT faces challenges to work with external intelligence it does not have. Till then it is a glorified search engine with extremely basic conversational skills baked into it. The OpenAI chatbot GPT is certainly not a fad and it is here to stay, but the expectations from the tool are sky-high and only time will determine whether it will bring the next technological revolution.
What are the limitations of ChatGPT?
ChatGPT’s core limitation includes sample biases, lack of contextualization, and lying (labelled as hallucinations), just to name a few.
Is ChatGPT free?
Yes, ChatGPT is a free-to-use web application. However, it has a paid version too with some additional benefits which costs 20$ per month.
What was ChatGPT trained on?
ChatGPT was trained on large text-based datasets from the internet. Some of the datasets are from the following websites:
3. Google Patents
5. Thrive Global
This is not an exhaustive list but it provides a general idea about the kind of data that went into the AI chatbot GPT.
What is the use of ChatGPT?
ChatGPT at this point is a niftier form of Google search, it provides quick responses to any prompts by the user.
Why does ChatGPT say network error?
ChatGPT network error on long responses happens because of server complications.
Can you get banned from ChatGPT?
Yes, OpenAI has terms of service guidelines which, if not abided by, can lead to a user ban.