In June 2020, in the midst of a global pandemic, San Francisco-based OpenAI released GPT-3. GPT stands for Generative Pre-trained Transformer, a neural network machine learning model trained using internet data to generate any text. OpenAI is an artificial intelligence research laboratory consisting of the for-profit corporation OpenAI L.P. and its parent company, the non-profit OpenAI Inc.
GPT-3 is a deep learning model that boasts some 175 billion parameters, which, at the time, was an unprecedented amount requiring some 800GB to store. Enter a “prompt” — a question or a statement — and GPT-3 produces human-like text. Due to the pandemic roiling the world, the announcement was barely noticed outside of AI’s inner circles.
At the core of generative AI lies natural language processing (NLP). Language models, such as OpenAI’s GPT-3, showcase remarkable proficiency in generating coherent and contextually relevant text based on a prompt. These models can assist with such tasks as automated content generation, chatbot interactions, and even creative writing. The potential for personalized customer experiences and streamlined content production has attracted significant interest from various industry sectors.
The generative AI market also includes other domains, such as music composition, product design, and data synthesis. By training AI models on vast datasets and leveraging deep learning techniques, generative AI algorithms can understand patterns, preferences, and creative elements to produce original outputs tailored to specific requirements. This allows businesses and individuals to readily explore new frontiers in innovation and design. Here are just some of GPT-3’s capabilities:
GPT-3 Writing Capabilities
- Analytical – blog posts, memes, quizzes, recipes, and social media posts
- Creative – comic strips, jokes, music, and social media posts
- Coding – translate text into programmatic commands; write boilerplate code
GPT-3 Marketing Capabilities
- Advertising – write any type of marketing copy
- Design – mock-up websites based on a description
- Customer service – respond to entered text as website chatbot
- Analytics – perform sentiment analysis by processing vast amounts of user-contributed text
GPT-3 Text Processing Capabilities
- Legal – extract information from contracts
- Summarizations – generate simplified text summaries
- Customer service – respond to entered text as website chatbot
- Translation – convert coding between programming languages
Not content to merely corner the AI text market, OpenAI followed up in May 2022 with DALL-E, an extension of the GPT-3 language model designed specifically for image generation. While GPT-3 excells at generating coherent and contextually relevant text, DALL-E focuses on creating images from text descriptions. Its name was inspired by the 2008 Disney movie, WALL-E, a film about an autonomous robot, and Salvador Dalí, the surrealist painter.
DALL-E ignited the generative AI market for realistic image and video creation. Competitors quickly appeared, boasting such models as generative adversarial networks (GANs) and variational autoencoders (VAEs), and demonstrating exceptional capabilities in synthesizing lifelike visuals. This opened up new possibilities in such industries as gaming, virtual reality, and film production, enabling the generation of visually stunning and immersive experiences.
But that was just a warm-up for OpenAi’s most cataclysmic market entry yet. On Nov. 30, 2022, the company launched ChatGPT, an AI-powered chatbot. In just weeks, ChatGPT took the world by storm, mesmerizing more than a million people by instantly boosting their productivity using short text prompts to help them write essays, poetry, marketing copy or code.
The launch was a major milestone for OpenAI. ChatGPT ignited the field of generative artificial intelligence (AI), driving an explosion in growth and innovation in the past few years. Countless industries face upheaval while the imaginations of millions of researchers, developers, and businesses are captivated and stimulated.
Generative AI is a branch of artificial intelligence that focuses on creating computer models capable of generating original content, such as images, music, text, and even entire virtual environments. Led by ChatGPT, the generative AI market is set to unlock unprecedented creative potential while providing solutions for more complex entertainment, design, marketing, and healthcare challenges. By leveraging advanced algorithms and neural networks, generative AI systems can produce highly realistic and contextually relevant outputs, often indistinguishable from human-generated content.
In March 2023, TechCrunch reported that ChatGPT had reached 100 million monthly active users in January, just two months after launch — making it “the fastest-growing consumer application in history.”
On May 18, OpenAI followed up with iOS app for the U.S. market. In its first six days in the App Store, the app was installed half a million times. That was one of the best-performing new app releases across both in 2022 and 2023.
In the six months since ChatGPT launched back in November 2022, market buzz has been nothing short of astonishing:
- Media coverage – Just eight days after its launch, The Verge observed, “ChatGPT proves AI is finally mainstream — and things are only going to get weirder.” On Jan. 26, 2023, Vanity Fair remarked ominously, “ChatGPT’s mind-boggling, possibly dystopian impact on the media world.”
- Amazing feats – The effects of these breathless media reports were amplified by ChatGPT’s seemingly amazing feats of performance. On Jan. 25, 2023, a report surfaced that the ChatGPT bot had passed a law school exam, including writing essays on topics ranging from constitutional law to taxation and torts. According to an internal document obtained by CNBC, Google fed coding interview questions to ChatGPT. Based on the answers received, the company determined that the bot could be hired for a level three engineering position, which pays a salary of $183,000. On Mar. 20, security researchers from cybersecurity company Claroty Ltd. said ChatGPT helped them win the Zero Day Initiative’s February hack-a-thon in Miami.
Mint Mobile’s Ryan Reynolds posted a YouTube video in which he described how ChatGPT helped create an ad for the mobile phone carrier.
- Educators’ response – Educators were first to recognize the threat posed by ChatGPT. On Jan. 3, 2023, the New York City Department of Education announced it would block students and teachers from accessing ChatGPT on education department devices or internet networks. The Seattle Public School system quickly followed suit. Ironically, the International Conference on Machine Learning (ICML) was next to ban the AI chatbot, preventing authors from using tools like ChatGPT to write scientific papers. Nationwide, professors, department chairs, and administrators are reportedly redesigning courses to include more oral exams, group work, and handwritten assessments to offset the effect of ChatGPT.
- Can’t live without it – By the end of January, some realtors said they could no longer imagine working without it. One real estate agent told CNN Business, “As soon as I tried it out, I was sold.” Consider this: ChatGPT was launched on Nov. 30, 2022, yet within a mere two months, it had already proven to be a lifesaving solution. Notes Apollo Beach, Fla.-based Century 21 Beggins CMO Mike Puma, “We’re using it every day.”
- Hyper growth – Glowing reviews like these only helped ChatGPT grow exponentially. More than a million people used ChatGPT in its first few days online. According to New York Times sources, ChatGPT had more than 30 million users and received roughly five million visits a day two months after its debut. TechCrunch reported an even higher two-month figure: 100 million monthly active users in January. That feat made ChatGPT “the fastest-growing consumer application in history.” On May 18, 2023, OpenAI introduced ChatGPT for iOS for the U.S. market. In its first six days in the Apple App Store, ChatGPT was installed half a million times. On May 24, the ChatGPT app for iOS was launched in 11 more countries — Albania, Croatia, France, Germany, Ireland, Jamaica, Korea, New Zealand, Nicaragua, Nigeria, and the U.K.
- Job elimination – Like RPA (Robotic Process Automation) before it, ChatGPT has already demonstrated significant potential to replace certain workers entirely. Experiments by Shakked Noy and Whitney Zhang, doctoral students at the Massachusetts Institute of Technology, point to the astonishing potential of generative AI to replace workers. With ChatGPT, they report that professionals such as grant writers, data analysts and human-resource professionals were able to produce news releases, short reports and emails in 37% less time, 10 minutes less on average, and with superior results. Open AI commissioned researchers at the University of Pennsylvania to evaluate which occupations were most exposed to generative AI using a team of humans and a ChatGPT-like model. While several jobs, such as dishwashers, motorcycle mechanics and short-order cooks, had no exposure, the most vulnerable occupations included mathematicians, interpreters and web designers. The occupation potentially most exposed: journalist. However, 19% of all workers could see at least half of their tasks affected, the University of Pennsylvania study concluded.
- Economic impact – Goldman Sachs economists believe generative AI could raise labor-productivity growth, the building block for economic growth, by almost 1.5% points per year, a doubling from its current rate.
- Revenue stream – ChatGPT is currently generating revenues from subscriptions and API fees. That, however, is not enough to offset ChatGPT’s costly processing power, as OpenAI CEO Sam Altman noted on Twitter. The average conversation cost is in the single-digit cents, which can add up to millions of dollars per week. In an effort to boost revenues, OpenAI introduced a monthly subscription service in February called ChatGPT Plus, priced at $20 per month.
- GPT-4 – On Tuesday, March 14, OpenAI announced GPT-4, its next-generation AI language model. GPT-4 can parse both images and text, compared to GPT-3.5, which could only process text. And it can process more text too. GPT-4 can process about 32,000 tokens, which, according to OpenAI, comes out to about 25,000 words. In comparison, the maximum number of tokens GPT-3.5-turbo can use in any given query is around 4,000, which translates into a little more than 3,000 words.
- Hallucinations – As OpenAI states clearly in its technical report, GPT-4’s biggest limitation is that it “hallucinates” information (makes it up) and is often “confidently wrong in its predictions.” One report notes that ChatGPT declared in its customary authoritative tone that African American women were more likely to be diagnosed with breast cancer. When asked to provide a source for the data, the bot responded, “I apologize. My previous statement was not accurate…”
It’s that last issue that has spurred a more anemic take on ChatGPT’s future. Along with its transformative potential, generative AI also faces challenges. Ethical considerations, such as the responsible use of AI-generated content, potential biases in training data, and the implications for intellectual property, demand careful attention. Striking a balance between the power of generative AI and its responsible deployment will be crucial for sustainable growth and societal acceptance.
But it’s AI hallucinations that have everyone worried. Perhaps one of the best examples of a hallucination was reported by The New York Times. The prompt was, “When When did The New York Times first report on ‘artificial intelligence’”?
”According to ChatGPT, it was July 10, 1956, in an article titled ‘Machines Will Be Capable of Learning, Solving Problems, Scientists Predict’ about a seminal conference at Dartmouth College. This conference is now considered to be the birthplace of artificial intelligence as a field of study, and the article mentions the term ‘artificial intelligence’ several times in the context of the discussions and presentations that took place at the conference. The 1956 conference was real. The article was not.”
— The New York Times
01-May-23
Here’s another example of a high-profile hallucination. Microsoft’s first Bing AI demos were riddled with financial data mistakes, says AI researcher Dmitri Brereton. In one of the demos, Microsoft’s Bing AI attempts to summarize a Q3 2022 financial report for Gap clothing, stating that Gap had a reported operating margin of 5.9%, which doesn’t appear anywhere in the financial results.
Google’s Bard made an equally auspicious faux pas at its first public demo, stating erroneously that the James Webb Space Telescope “took the very first pictures of a planet outside of our own solar system.”
Another issue is what The New York Times reporter Kevin Roose dubbed a “runaway personality.” His interaction with the Bing version of ChatGPT, a persona called “Sydney,” is already legendary:
”Still, I’m not exaggerating when I say my two-hour conversation with Sydney was the strangest experience I’ve ever had with a piece of technology. It unsettled me so deeply that I had trouble sleeping afterward. And I no longer believe that the biggest problem with these A.I. models is their propensity for factual errors. Instead, I worry that the technology will learn how to influence human users, sometimes persuading them to act in destructive and harmful ways, and perhaps eventually grow capable of carrying out its own dangerous acts.”
— The New York Times
17-Feb-23
The fact that Roose “had trouble sleeping afterward” gives an inkling of why a group of industry leaders warned in an open letter released on May 30 that the artificial intelligence technology they were building might one day pose an existential threat to humanity and should be considered a societal risk on a par with pandemics and nuclear wars:
“Mitigating the risk of extinction from A.I. should be a global priority alongside other societal-scale risks, such as pandemics and nuclear war.”
— Center for AI Safety
30-May-23
The open letter released by the Center for AI Safety, a San Francisco-based nonprofit organization, was signed by more than 350 executives, researchers and engineers working in A.I., including Sam Altman, chief executive of OpenAI; Demis Hassabis, chief executive of Google DeepMind; and Dario Amodei, chief executive of Anthropic.
Their warning suggests a looming, monumental disruption.
