What's the big deal with chatGPT?
Almost exactly two weeks ago, OpenAI released ChatGPT, a powerful new chatbot that can communicate in plain English. ChatGPT is a variant of GPT, which stands for Generative Pretrained Transformer, which is a type of large language model (LLM) developed by OpenAI. These LLMs are algorithms that are trained on a large text dataset (books, social media content, blog posts, etc.) to “learn” and predict the next word in a sentence. It could be thought of as something similar to the autocomplete on your phone, but on steroids (tons of it). While systems like this have existed before (Google Duplex is a good example), the quality of outputs was much lower than that produced by an average human. This new model is however much better, startlingly so!!!
Put simply, ChatGPT rewrote all the rules. If individuals and businesses understand the significance of this change and proactively adapt, they will be at a considerable advantage. Especially since this is the beginning of an entirely new approach to working with knowledge (generation, curation, and transfer). ChatGPT is just the first of its kind, and many similar tools are going to be available shortly with their capacity improving exponentially. This is a very big deal, and those who act on this first will be at a considerable advantage.
For those who try out ChatGPT for the first time, it would look like a novelty item, and on a technical level, it’s not profoundly different from other AI systems. It’s just better due to being trained on a larger, more filtered dataset. Also, when you look at articles about ChatGPT, you mostly come across niche examples like writing poems or children’s books, offering recipes based on items in the fridge, etc. These examples don’t show what the system is capable of. Further, AI has notoriously been overhyped in the last several years. We have had Elon Musk promising AI-assisted self-driving for nearly a decade without delivering on the promise. AI is also overhyped by others, for instance in 2019 Sundar Pichai famously demonstrated how their AI-driven chatbot Duplex could function as a personal assistant and also talked about how the system passed the Turing test, only to be later proven wrong. Mostly, AI systems have thus far only shown their potential, but with this system, I believe we have crossed the rubicon into a place where AI actually becomes widely used.
ChatGPT is free to use and is holding up against a large number of people using it for a wide variety of general use. Until now, most AI systems have been tailored for niche applications where the cost of failure was high. Now, it’s finally being applied to mundane tasks like copywriting, picture creation, summary generation, etc. The advantage is that the system can now get a massive amount of feedback to improve itself. And, the system is being used a lot. In a relatively short one-week time frame, it got more than 1 million users, making it probably the fastest-growing platform in history. To put this in context, Netflix got to 1 million users only after 3.5 years.
All this talk sounds great but what are some examples that the system actually enable,
While the system was designed to enable sophisticated language (English, Hindi, Mandarin etc) operations, it can also work with computer languages. To give you an idea of what this looks like, I asked chatGPT to create a tik-tack-toe game. While this looks relatively easy, it would have still taken a human about 20 mins to code it up (assuming they were decently knowledgeable), but chatGPT did it in under a minute.
I also tested the code, and it works!
Second, it has an incredible capacity to perform different kinds of writing with more significant implications than might be initially apparent.
Marketing, advertising, consulting, journalism, publishing, and finance, where communication with clients and stakeholders with high-quality written materials is essential will never be the same.
Third, rather than using this system to give answers one could use it in a collaborative, hybrid format because of it being designed with conversation in mind (You can see this in the tik-tack-toe example above). Instead of prompting an AI and hoping for a good result, humans can now guide AIs and correct mistakes. This allows the human to treat the AI as a partner and ensure errors are minimized while maximizing output in minimal time. This means experts will be able to fill in the gaps of the AI’s capability while the AI takes over the more boring tasks. This mode of interaction led to players of Go, as well as chess, to improve their performance as the AI system that beat them also taught them about moves/strategies they couldn’t have humanly figured out themselves. From an educational perspective this could be fabulous.
While these capabilities are nothing short of amazing the reason why I think we have crossed the rubicon, into a world where AI will be all around us, is because at long last the larger public will be playing with AI directly. They are discovering ever new ways of using it and I am certain several new unicorns are going to be built merely by fine-tuning these models. We have already seen a couple of really great startups (copy.ai, Jasper) build on the basis of just GPT-3 but the limits of these tools have not even been defined yet.
Everything is not however peachy and awesome as one might think. The reality is that there are still some fundamental problems with this kind of AI. The main one is the extremely convincing sounding answers that the tool is able to give even when conveying complete nonsense. It also can’t explain what it’s doing or how it constructed the information that it is providing (no referencing capabilities). Lastly, it’s currently hard to have nuanced conversations with it about controversial, or sensitive, subjects. This last issue is mainly due to restrictions placed on the system to avoid the propagation of systemic biases as well as unethical opinions that might exist within the dataset that was used to train the model. I am certain that in the coming weeks, months and years, these issues are going to be become more acute as the usage of the tool spreads and we will have no choice but to deal with the consequences, just like how we dealt with the consequences of social media propagation.
These shortcomings are not unique to AI-driven systems and one way in which I believe they are going to be resolved is through AI-human hybrid work. Writers don’t have to write articles alone, coders don’t have to program alone, students don’t have to learn alone, teachers don’t have to teach alone, the list goes on. These are types of collaboration that didn’t exist even a month ago but I suspect this is going to be such an integral part of our lives that within a few months this would be the new normal. This tool along with remote work the boundaries of jobs have completely shifted. I am not going to claim that I know where things are going to go but it’s going to be a whole new world. Do note that this is but the first model of it’s kind that has become public, it’s highly likely that in the coming months there are going to be similar products released by google, meta as well as others which might be faster, better or with new capabilities that have not even been defined yet.
I’ll leave you just with this quote and one final thought.
The real danger is not that computers will begin to think like men, but that men will begin to think like computers.
Sydney J. Harris
So, please do yourself a favor and start experimenting with AI and have discussions about its implications, we are merely at the tip of the iceberg, and a whole lot more is coming down the pike.
Disclaimer: The views and opinions expressed here belong solely to AG and shouldn’t be attributed to any organizations he might be associated with.