(read time: 10 mins)
- Why should we care?
- What is chatGPT
- Concrete examples
- Should we worry?
- Potential headwinds
- What should I do now?
- Disclaimers
Why should we care?
If you’re reading this, there’s a strong possibility that you have already encountered ChatGPT. You may have used it, seen someone mention it on social media platforms such as Twitter, YouTube, TikTok, or Instagram, or encountered it in the media through TV, newspapers, or magazines.
Before delving into what ChatGPT is, I’d like to emphasize why it matters. I don’t purport to have all the answers, but I believe it represents a new era where we are witnessing magic through technology. The most recent and obvious examples of this are the widespread adoption of smartphones and, prior to that, the internet. These technologies have become so ubiquitous that we may have forgotten just how magical and awe-inspiring they once seemed.
Interestingly, it’s not the magic that’s the true breakthrough here. We’ve transformed the way we live our personal lives, businesses have adapted their operations, and our roles as employees and business people have evolved. There are also countless underlying changes that occur behind the scenes when using these technologies, and often the best uses of technology are those that go unnoticed.
Some people are passionate about technology, both for its own sake and for the opportunities it affords, while others just want to get on with their lives, pursue their interests, and generally live a fulfilling existence without worrying about the technical details.
One measure of the success of a technology or product is the speed at which it is adopted and shared, and the trend is towards faster and wider distribution:
We are still in the early stages, but plenty of people feel this is a pivotal moment. This wouldn’t be the first time a technology is over-hyped (an often criticiszed but still interesting view is the Gartner’s hype cycle of technology -wikipedia-). Indeed, many of the predecessor components of chatGPT (and other related technologies) have been hyped in the past. If you measure by consumer adoption, some could say that they didn’t meet the inflated expectations, but I would argue that the work happening in the background paved the way for what we are experiencing now.
So, to end the section, why you should care?
- It looks like magic
- Adoption is growing faster than most previous technologies
- There is a good chance that this technology (directly or through some derivatives) will again affect our daily lives and how organisations operate
- In particular, there are obvious huge implications to the fields of Education and Health, which are two central elements of a prosperous society that currently feels for many that is not matching the degree of progress in other areas
What is chatGPT
UPDATED: Since publishing this post the following article came out and I recommend everyone with strong interest in the topic to take a read. It’s long (85 mins) and it has some maths, but it’s very clear and several deep parts can be skipped and you’ll still get most of the value from reading it: What is chatGPT doing… and why does it work? – Stephen Wolfram
ChatGPT is a web application hosted on chat.openai.com that was launched on November 30, 2022. It features a straightforward user interface, resembling a chat or direct messaging experience, with the key difference being that there is no human on the other end. All the text is generated by a computer server.
While this type of interface may seem familiar, with similar experiences offered by companies as customer support, it often leaves users frustrated when they are unable to find solutions to their problems, and they try to find ways to get a human agent to solve their issues.
What sets ChatGPT apart is its ability to produce text that is well-written and sounds like it was created by a human. However, it goes beyond this, as people have already discovered use cases that go beyond what a human could realistically produce in seconds.
ChatGPT was developed and published by OpenAI, a non-profit organization based in the United States. To use the application, you need to sign up and create a user account. At launch, it was made available through a waitlist and users were invited gradually.
Using ChatGPT involves visiting the web application and sending messages, just like you would to a friend or family member. You can even have minor typos and it will still work. After sending a message, you’ll receive a response that is gradually typed out on the screen. You can either wait for the response or force it to stop, but currently, you cannot send multiple messages in quick succession.
Mastering the use of chatGPT requires a bit of experimentation to discover its potential applications. The outcome of your interaction with the application can be greatly impacted by even subtle modifications to your inputs, commonly referred to as “prompts”. As a result, some individuals have coined the term “prompt engineering” to describe the art of crafting the ideal prompts for optimal results. An alternative term, “prompt craft,” has also been proposed and I actually think it reflects better the process of getting better at it.
Concrete examples
There are thousands of interesting use cases for this technology, this is why it’s great to try on your own and dare to be creative. Also by watching what other people are doing you can draw some inspiration.
- Improving language. This post for instance has ben enhanced through ChatGPT. I am not a native english speaker and tend to be very verbose when I write and have typos. The post may still have some of that verbosity or typos, and that remains my fault 100%!
- Creating texts “in the style of”. An internet favourite is to get stories like if they were a Sonnet by Shakespeare. I will mention in a separate post, but this use case was actually made public earlier by a different product from OpenAI, that actually is strongly related to ChatGPT
- Translating between languages
- Create workout routines, meal plans, tour guides of cities
- Create essays on diverse topics (creating a big challenge for certain forms of education!)
- There is a ton more, some more successful than others (example it can “play chess” by using the standard notation, but it will often make illegal moves!), I’ll focus on more use cases in separate posts
Many of these use cases you would have carried them with a Google search (plus some work), that’s why currently there is an ongoing debate regarding the impact that this technology will have on web search (googling, or Microsoft would expect, binging?)
Should we worry?
There are many risks associated with this technology, both if things go nice and well, but also if things take wrong turns.
On one side of the spectrum, there have been people arguing and researching that aiming for a super intelligence (often called AGI – Artificial General Intelligence) pose a catastrophic risk to humanity. One of the main voices on this camp is Eliezer Yudkowsky and many others that even founded a non-profit called MIRI (Machine Intelligence Research Institute) that has as one of their main research programs to study how we can keep AI in safe boundaries.
On a friendlier side (vs. humanity extintion I mean), if some of the scenarios where the technology ends up showing a lot of its potential, millions of jobs will be disrupted, and those workers may find challenges to acommodate to the new world. This isn’t new, and has happened with every technology revolution, but it doesn’t mean it won’t be a societal problem (it will be a problem for those that end up getting directly affected). I am a positivist technology, and believe that we as society need to do changes to make sure these technological progresses continue happening and find ways to reduce the downside effects on society (or parts of it). I am definitely not in the camp of curtailing innovation and progress.
Which jobs would be threatened? There is still a lot of debate and we still need to see how things evolve, but the initial thinking is that this will mostly affect knowledge workers. Some people claimed that software engineers are at risk, as one of the areas where chatGPT (and some of its cousins, to be talked about in other ocassion) is showing a lot of promise, but actually I do believe that in general this will be a tool that will make people more efficient and effective at what they do. Indeed, one of the current issues with chatGPT is that it can often write false or incorrect things, so having some knowledge and experience is quite important to get value from it (specially if the cost of getting something wrong is high!).
Another important element is associated to getting flooded by content, by swaths of text created by these tools. It will only make the issues of “fake news” worse. Also, as we’ve seen with other applications of algorithms, there can be biases of all sorts (bias is actually a formal statistical term, besides the colloquial use) that will reflect mainly the bias present in the data used to feed the computer program that answers.
Potential headwinds
Regulation and government intervention may soon impact the field. I think this is one of the reasons that Google seems to be going slower –What will be the impact of chatGPT on Google (a slim response)– in going public with solutions to this. Given how the tools were created (trained), they can often write back with sensitive responses, even aggressive, racist, discriminatory, dangerous, etc. Indeed, OpenAI is currently using a model (“GPT-3.5”) for which training has finished some time ago, but they do work on the inputs and the outputs for the model to apply what they call a “content policy” to avoid or at least reduce the chance of getting content for which they wouldn’t like to appear in the cover of major media outlets.
Reaching the natural limits of the current architecture. The underlying language models are based on an architecture that may not allow it to overcome the main current drawbacks. In a VERY simplified way, the models currently know which sequences of words are plausible, and thus use an algorithm to decide on the next word, each at the time (I am simplifying a few bits here, but conceptually should work). For reference, the field of “AI” has had what has been called summer and winters (we are currently on a hot summer!), and the cycle may turn back again
Cost implications and lack of business model. There are several conversations happening that compare the cost of running a typical web search vs. running these conversations through Language Models, and the differences are likely a couple of orders of magnitude. For sure there will be improvements in the next months or years, but it will definitely be somehow expensive, so I don’t think it’s obvious that we’ll get free access to these tools for much longer (yes, we are currently part of the product and provided implicit and explicit feedback when using the tool). Anyway many companies racing to get new breakthroughs and use cases, so I remain optimistic that things will eventually reach a point where there is wide access to these tools in a sustainable way.
What should I do now?
It depends (former management consultant here). But at the minimum I would try to sign up to ChatGPT, the new Bing and Google’s Bard). There are other platforms, I haven’t tried many, but had some fun with character.ai.
If you want to go deeper, start to follow people on twitter, I have some general recommendations of accounts in another post, which are not specific to this topic, but most of them are talking about this anyway.
Depending on your background, intentions and available time you can also get a bit closer to AI/Machine Learning/Natural Language Processing, there are many free courses, in platforms like Coursera, Udemy, Udacity, even several universities have websites for their courses, often with all the materials (and video lessons) available. Then specialist organisations like fast.ai, deeplearning.ai, and I am sure missing tons, but there is plenty of free resources, the most expensive thing is the time, attention and consistency to do this (well, if you get very serious on this, the cost of the machines to run these models, but it won’t be a problem per se, specially if your goal is not to create an alternative to chatGPT from scratch -that would be a very very very tricky and laborious effort-).
Disclaimers
Disclaimer: I do not possess a substantial stake or financial involvement in any company mentioned in the post or currently working directly in this field. I am an enthusiastic advocate for technology and its potential to improve the human experience. I acknowledge that advancements can be unevenly distributed, but I remain optimistic that obstacles to benefits are decreasing and we have yet to discover ways to accelerate progress for a wider reach.
Disclaimer 2: Several parts of this post were first written entirely by me, and then improved with chatGPT through the following prompt: “please improve this language: [several paragraphs at the same time]”, I am not getting yet a consistent experience here, I need to practice the craft
Disclaimer 3: the main image of the post was a quick and dirty generative image created with playground.ai (after a few failed attempts with DALL-E from OpenAI. Image generation is another booming area currently, and also getting the right images requires improving how prompts are generated, this was my last try: “A laptop sitting on a table inside a coffe house. In the screen you can see a chat conversation with several exchanged messages. The most recent messages have an icon with an avatar of the two people talking, one is a computer chip an the other is just the shape of a human head with no details. The logo of Open AI in the background of the screen”.
juan M 
Leave a reply to Juan Cancel reply