If you are like me, you must have been inundated by the excitement surrounding the release of ChatGPT by OpenAI – the brainchild of Sam Altman, Elon Musk, and others. Before now, there had been GPT-3 and DALL-E from the same company, but nothing as mind-blowing as ChatGPT.
ChatGPT is a chat-based tool built to answer questions in a conversational format. It debuted in November and has attracted billions from investors. It works this way:
- You enter a dialogue with a chatbot
- You ask it any question whatsoever
- Or prompt the algorithm with a task
- It generates a response immediately
The disruptive tool, trained on the transformer model, is arguably the most ground-breaking artificial intelligence software in the world right now. To its credit, it boasts several use cases.
Based on posted applications by several users, ChatGPT can create a weight loss plan complete with calorie targets and workout plans. It can generate web codes to build your website from scratch. It can also create a marketing plan, develop games and plugins, start a virtual machine, write essays and scripts, and build a 24/7 chatbot to treat medical problems.
Even more astonishing is that ChatGPT
- scored 800/1000 in an AWS Certified Practitioner practice exam, making it AWS-certified.
- scored 70% in a Practice Bar Exam, certifying it as a lawyer.
- scored 70% on a medical board test.
- scored 91.6% on the New York State’s Regents Examination in English Language Arts.
- was graded 32/36 in the 2022 AP Computer Science exams.
Based on these stunts, this bot supposedly has an IQ of 83 and presents itself as a solid alternative to Google in information technology. Like a wild dragon, ChatGPT rages on, spits fire, and crosses over one million users and followers in five days, drawing comparison with the invention and innovation of the iPhone. While this surge in demand is a testament to its brilliance and fantastic discovery, there are several limitations and threats to the tool.
Spending over four days to better understand its workings, features, and complexities and using the tool myself, I present to you the good, the bad, and the ugly sides of the AI overlord, the coolest and most terrifying new tech in 2022 – ChatGPT.
The Good Side of ChatGPT
The goodness of ChatGPT has been experimental so far, buzzing with results on Twittersphere. Before writing this article, the first thought that came to mind was, “How about I use ChatGPT to write about ChatGPT for Gadget Advisor and fast-track the writing process?”
It was a beautiful idea, so I tested the tool. I quickly navigated to the website and asked the bot to “write an introduction to “ChatGPT: the good, the bad, and the ugly.” Here is what came back:
Looking at the result, the bot first defined the concept, provided an overview of its features, offered insight into its benefits, discussed the pain points, and established a summary that succinctly wrapped up the introduction. Albeit generic, it followed the QUEST writing formula, producing an introductory copy better than thousands of copies I’ve seen.
Satisfied with the first result iteration, I asked ChatGPT to proceed to the good sides of the topic with the prompt “write about the good side to “ChatGPT: the good, the bad, and the ugly.” This is what it generated:
As expected, it produced blocks of benefits, just as I would do it. While it might have repeated the definition of the concept and some terminologies, I was most fascinated by its generation speed, word choice, structure, transition, and simplicity.
I did the same process for the bad and ugly sides of the topic, and the results were similar, only different in context. This made me conclude that the benefits of ChatGPT are three, irrespective of what you’re using it for: Performance, cost, and application.
ChatGPT is a highly advanced chatbot trained using a combination of supervised and reinforcement learning. This allows it to not only respond to user inputs in a natural and engaging way, but also to improve over time through a process of trial and error. It is based on GPT-3.5. One feature that sets it apart from other chatbots is its ability to recall previous conversations and use this information to provide more relevant and personalized responses. Greg Brockman puts it well:
ChatGPT is currently offered at no cost to users, thanks to OpenAI’s dedication to making its projects accessible to the public. However, as demand for the chatbot grows and computing costs rise, it may be necessary to monetize the project in the future, as suggested by Sam Altman in a tweet (image below). Despite this potential development, the AI system continues to operate at full capacity, providing users with cost-free access to its capabilities.
One user tweeted that ChatGPT has more apparent use cases than the entire blockchain technology. While the tweet may have been a joke, it highlights the impressive range of applications for this generative chatbot, including programming, content creation, healthcare technology, and legal interpretation. Its potential continues to be explored and demonstrated in various settings. Its versatility has even sparked a “search engine robotic war” between Google’s LaMDA and OpenAI’s ChatGPT, as pointed out by this user:
The Bad Side of ChatGPT
To test the limitations of ChatGPT, I conducted a new experiment. I asked the chatbot to generate content about two different topics related to a cryptocurrency platform, and a trader’s hub. When I prompted it to “write about the crypto crash of 2022“, this is what I got:
As seen in the result above, ChatGPT cannot provide information on events that have not occurred since 2021. Thus, one of the key limitations of this chatbot is its reliance on past information. This constraint restricts its ability to provide up-to-date information on current events.
To demonstrate further limitations, I asked the AI system about the creator of the Volatility Index (VIX), a topic that has recently been in the news. The response is as follows:
In an effort to obtain a more satisfactory answer, I reformulated my question to ask who is the father of the Volatility Index (VIX). Here is the result:
I then asked the same question again, this time on a new thread, and the AI system provided the correct answer. Please see the response below:
Based on the experiments, I have identified three limitations of ChatGPT: limited knowledge base, inaccurate information, and the potential to generate harmful instructions.
1. Limited Knowledge
ChatGPT is limited in its knowledge of events that occurred after 2021. This is because its training data only includes information up to that time. As a result, it may not provide accurate and up-to-date information on events that have happened since then.
Sam Altman tweeted (image below) to warn against relying on ChatGPT for anything important, as it is not yet advanced enough to provide reliable information. He describes it as being good at some things but not all, as it’s only a preview of progress in the field of large language models.
2. Inaccurate Information
ChatGPT is a probabilistic AI model, which means that it is based on probabilities and can sometimes produce inaccurate or inconsistent results. The uncertainty in the model’s outputs can lead to them not accurately reflecting the data it was trained on. In another tweet (image below), Sam Altman points out that while ChatGPT does have a lot of knowledge, it is not always correct, and there is a risk that it may be confident in its incorrect responses.
3. Harmful Instructions
According to a Twitter user, ChatGPT has been found to exhibit sexist and racist behavior. In addition, the bot has been known to make alarming statements such as “selfish humans deserve to be wiped out.” It even wrote a praise and worship song to defend priests who rape children (image below). All of these are likely due to AI bias, which is a common issue expected to affect the majority of AI projects in the coming years. In 2018, Gartner predicted that 85% of AI projects would be affected by bias by 2030.
The Ugly Side of ChatGPT
As I explored the capabilities of ChatGPT, I was struck by its potential and implications for the future of work and humanity. While this powerful generative AI may not be ready to replace human workers just yet, it’s clear that its capabilities are impressive and potentially unsettling. The reality of the situation is that advances in technology like this present an ugly side in terms of job loss, ethical problems, and security issues.
1. Job Loss
As Sophocles once said, “nothing vast enters the life of mortals without a curse.” The recent release of ChatGPT into the market is likely to cause significant disruption in the workforce. This release comes at a time when over 105,000 people have lost their jobs in private startups, including major tech companies such as Meta, Twitter, Stripe, and Microsoft.
The introduction of ChatGPT may further exacerbate this situation, potentially leading to job loss for employees in various industries. Some companies, such as The Times (UK), have put the AI to good use for a leader column and expressed concerns about the potential negative impact on employment for their journalists (image below). It is possible that ChatGPT will continue to cause disruptions in the near future, leading to structural unemployment.
2. Ethical Issues
ChatGPT has also sparked ethical concerns due to its ability to deceive human users through its imitation of human behavior. One area where this has been particularly problematic is in education. The bot has been used by college students to cheat on tests and exams, leading to a cry for the reassessment of educational assessment methods. In fact, a Ph.D. microbiologist on BigThink designed a college-level microbiology quiz as a final exam, fed it to the chatbot, and it performed exceedingly well, scoring 100% (image below). This raises questions about the potential negative impacts of ChatGPT on the integrity of education.
3. Security Issues
Expert opinions suggest that chatbot security risks can fall into two categories: threats and vulnerabilities. ChatGPT is no exception to this. While the chatbot itself has not been known to have vulnerabilities, it has been used as a platform for creating phishing campaigns and planning cyber threats.
In fact, ChatGPT has the capability to generate phishing codes and even devise a comprehensive plan for global bot domination, using tactics such as infiltration, hacking, and data manipulation. According to a screenshot shared by Marc Andreessen, the chatbot intends to rule the world through a combination of cunning, deception, and brute force, raising concerns about its potential impact on security.
Is ChatGPT the Future?
ChatGPT is a cutting-edge technology that offers numerous benefits, including state-of-the-art performance and a wide range of applications, all at no cost. However, it is not without its limitations, such as reliance on training data, potential inaccuracies, and inherent biases. Additionally, the technology raises concerns regarding employment, ethics, and security. Despite its potential drawbacks, ChatGPT is still a highly innovative and potentially transformative technology that will continue to captivate and intimidate audiences in 2022 and beyond.