I. ChatGPT is here. What hath OpenAI wrought?
It’s been over six months since the tech firm OpenAI publicly released ChatGPT into the universe.
And our world will never be the same.
Hopefully, that last statement doesn’t seem like hyperbole. But if it does, that likely means that you haven’t yet taken a deep dive into ChatGPT or some of the many other generative AI tools that have made their appearance over the last few months.
Everyone I know who has spent significant time with these tools has indicated – when they’ve come up for air – just how powerful and astonishing they are. The consensus is that it seems almost impossible to exaggerate the importance of what we’re witnessing.
In fact, back in 2016, Amazon founder Jeff Bezos said, “It’s hard to overstate how big of an impact AI will have on society over the next 20 years.”
And yet, here we are – seven years later – and it feels as if the world has somehow been caught off guard, both by the sudden appearance of ChatGPT and by the tsunami of change that it’s brought forth. This is despite the fact that we’ve seen many uses of AI creeping into our lives over the last few years, all from tech’s largest and most well-known companies. Ranging from Google’s autocomplete feature in Search, to Microsoft Word’s red squiggles indicating misspellings, to Apple’s Face ID facial recognition system on the iPhone, these features have quietly become part of our lives. Even so, none of them has had a fraction of the impact of ChatGPT.
Why is that? Likely because these earlier appearances of AI seemed simple and convenient, and they didn’t appear to herald a new era.
And yet, with the appearance of ChatGPT, the world has, in fact, changed. The AI Era has officially begun, bringing with it a flood of hype, a rash of new AI-focused companies and the requisite gold rush, in addition to much gnashing of teeth and a significant number of predictions of doom.
Many of those predictions of doom include massive job losses. In fact, earlier this year, Goldman Sachs suggested that 300 million jobs worldwide could be lost or “diminished” because of AI.
So, how much teeth-gnashing is appropriate here?
II. ChatGPT and generative AI beg the question: Where is all this headed?
When we imagine the continued change we can expect over the next couple years from ChatGPT, its future iterations and competitors, and all things AI-related, it would be difficult to argue with Bezos’s words about the outsized impact that AI will have on society, business and life.
In fact, taking it even further than Bezos, Sundar Pichai, CEO of Google (and its parent company, Alphabet), has said that “AI is one of the most important things humanity is working on. It’s more profound than electricity or fire.”
Now, that’s a bold assertion! And it should be sufficiently head-spinning to make even the most passionate advocate of AI pause for a minute and say, “Really? More profound than electricity? More profound than fire?”
But what if we began to see AI as something that literally could be integrated with our brains to help us make better decisions in our jobs and our personal lives? While that may seem like a stretch for now, might our current ability to access AI on-the-fly be the start of something really big? Might it allow each of us to become superhuman? Or at the very least, hyper-intelligent? If so, the comparison to the discovery of fire begins to make some sense.
Happily or unhappily, we’re not there yet. And I’m not sure that we’re just around the corner from what’s called the Singularity. But it is remarkable to see that many of the sharpest minds in the business – those who’ve been working on AI the longest and who understand it best – do not mince words when it comes to the significance of what AI will mean to all of us.
III. So, how head-spinning is the current rate of innovation and change in generative AI?
In many ways, the rise of AI feels a lot like the “this is really huge” excitement that followed the introduction of the Netscape Navigator web browser back in 1994. However, one big difference from 30 years ago is just how quickly ChatGPT (and generative AI overall) are growing better, stronger, faster, seemingly by the day. Yes, we thought things were changing quickly in the early days of the web, but in comparison to ChatGPT, the early web was a slow-moving train, while ChatGPT appears to be traveling at supersonic speeds.
One way of quantifying that speed is to look at how quickly ChatGPT went from zero to one million users. From its launch date, that journey took a mere five days, which makes it the fastest growing online service of all time — at least until then. (That record stood only until July of 2023, when Facebook parent Meta’s new, would-be Twitter-killer, Threads, passed one million users in just a few hours.)
And then, incredibly, after only two months, ChatGPT had 100 million users. For some context, TikTok took nine months and Instagram took two and a half years to get to 100 million. And, seemingly a lifetime ago, Google and Facebook each took about five years to reach 100 million users. Just in the last month, Threads made it past 100 million users after only five days, which is an indication of two things: the market power of Facebook/Meta and the desire of millions to find an alternative to Twitter.
But it’s not just the adoption rate of ChatGPT that’s been off the charts; since its original release, the improvement in the tool itself has been dramatic. At its launch last November, it was officially labeled model 3.5, and not even four months later, in mid-March, OpenAI introduced ChatGPT-4. This newer model features greatly expanded and updated data, fewer errors, and the ability to synthesize much larger inputs of data.
Note that it’s necessary to subscribe to OpenAI for $20 monthly in order to have access to ChatGPT-4, but I would suggest that’s a very small and reasonable expense, given the much-improved outputs, in addition to having priority access to the tool during its busiest periods.
But why should this record-setting adoption rate and the rapid improvement in the tool be important to you? Because hitting 100 million users so quickly – as well as seeing how it’s already had a big upgrade and has led to the launch of so many new tools, both complementary and competing – are powerful signs that not only has generative AI quickly become a key part of the online environment, but it’s increasingly an important part of many people’s lives.
To that point, if you haven’t yet felt generative AI’s impact on your work and life, just wait. You will. And it may happen a lot sooner than you think.
IV. How should you think about – and work with – ChatGPT?
Over the last several months of working with and learning about ChatGPT and its competitors, I’ve become aware of — and developed — quite a few rules of thumb regarding best practices. Here are three of the most important:
1. Do not put personal or confidential information into your ChatGPT queries (known as “prompts”). It’s important to understand that all generative AI tools get smarter by drawing upon – and learning from – the information contained in the prompts, so using and cataloguing whatever information you put in your prompt is, by definition, how they operate.
However, given how new and different this tool was, soon after ChatGPT was released many people raised reasonable concerns about the privacy and confidentiality of the information contained in their prompts. Because of the uproar, in late April OpenAI made opting out of sharing this information much easier by providing a simple toggle switch in the Settings.
And yet, even if you do choose to opt out of sharing your chats with OpenAI, I would still shy away from using any personal or confidential info about you or your clients. At this point, doing so would seem foolish – and there are other, much better ways of getting the information you want out of ChatGPT without taking that risk. In a nutshell, you can include relevant information in your prompts without it being personally identifiable information. I’ll go into this in more detail in future posts.
2. You must fact-check all responses you receive from ChatGPT. Never assume that your beautifully written responses from the chatbot are not filled with errors. And yes, even about factual matters that you think would be the lowest-hanging fruit for it to get right.
As I described above, things have gotten better in that regard with ChatGPT-4 (which, again, is well worth the money, even if you don’t use it every day). But buyer beware, because you’re still not getting any guarantee that ChatGPT’s responses are completely accurate. Even though ChatGPT’s fact-deprived “hallucinations” have become less frequent, they have not gone away entirely.
Whatever you do, learn from the mistakes already made by others. I’m thinking about New York attorney Steven A. Schwartz, who used ChatGPT to draft a legal brief in a personal injury case without taking any steps to check its accuracy. Give him credit for being an early adopter willing to try out a new technology, but then take all that credit away because he obviously didn’t do any due diligence on how the technology works. Unfortunately for him, the brief included several entirely made-up cases that he cited as precedents.
Understandably, the judge in the case was not pleased, and as The New York Times reported, Mr. Schwartz faced sanctions. His argument at the sanctions hearing was, in essence, that “I did not comprehend that ChatGPT could fabricate cases.” Not surprisingly, that did not go far with the court, which sanctioned him and another attorney $5,000 each, writing, “Many harms flow from the submission of fake opinions.”
With all that in mind, ChatGPT should be seen as a hard-working personal assistant – but unfortunately, one whom is both overconfident and untrustworthy. You can definitely save yourself many hours of time by being a savvy user of ChatGPT, but all savvy users of the tool know to treat its responses as rough drafts that must be carefully reviewed and vetted.
3. For any critical content that you want to make public, you’ll want to proofread, personalize and polish the responses you get from ChatGPT. (Just to be clear, this is in addition to the mandatory fact-checking described in #2.)
Think of ChatGPT’s responses as simply a starting point for you, and not as final, ready-for-prime-time versions. My suggestion here is to go beyond simply making sure the content is factual and readable, and to further craft it so that it’s also compelling to read, and so it has the best chance of having the impact you’re seeking.
Why is this important? Because, by its very nature, ChatGPT is designed to produce fairly generic, even vanilla output. It may be well-written grammatically, but don’t let that fool you. It’s essential to remember that generative AI programs are not actual writers, they’re machine learning models. That means they crunch through huge amounts of data and respond with the most likely next word based on nothing but statistical probability.
Now, given that ChatGPT is a statistical probability machine, it is incredibly impressive in its writing. But as we’ll discuss in Part VII, when you want your writing to be substantive, personal and compelling, you’ll want to continue to work on what ChatGPT gives you by proofreading, personalizing and polishing it.
Because if, in fact, you’re looking to open minds and wallets, you’ll want your writing to stand out and rise above, and not just subsist at the basic level of generative AI. ChatGPT and the rest will raise the bar for what’s considered minimally acceptable writing, but that simply means that if you want to get your writing – and most importantly, what you’re trying convey – noticed, you’ll still have to do better than the minimum that ChatGPT provides.
Ultimately, it’s critical to remember that while ChatGPT is fully capable of writing some of your content, it’s humans who will be reading all of it.
V. How can someone make sense of all the many generative AI tools flooding the market?
That’s quickly becoming an essential question to understanding where and how all the various flavors of generative AI fit in, both in terms of each other and in terms of AI users.
A good starting point would be to know that ChatGPT licenses its API (Application Programming Interface) to developers around the world. By doing so, they can produce and sell their own, often narrowly focused, generative AI products, all while relying on ChatGPT as the AI powerplant. Of course, just having ChatGPT behind the scenes is no guarantee that the organization that’s licensing the generative AI will produce a worthy tool. Buyer beware, as always.
Drilling down somewhat, Hitendra Patil, who has long been an influential voice regarding accounting technology and accounting firm innovation, suggests that there are (or will be) three main ways AI will affect the accounting profession, and I think it’s also a useful model for looking at what’s currently taking shape throughout the worlds of business and law.
In his article, Three Ways A.I. Will Take Shape in Accounting, Patil asks his readers to consider three main categories of artificial intelligence: public, private and custom:
Public AI, which includes ChatGPT, is available to anyone and can be used for relatively basic, general tasks that are not specific to a single profession or industry. Perhaps not surprisingly, public AI will remain both less accurate and less expensive to the user than the other two categories.
Private AI, which would include BloombergGPT, is being developed by large organizations that have access to the data of thousands of companies, millions of individuals and billions of transactions. Often built with a specific industry or profession in mind, private AI will be more expensive than public AI, and is likely to be more accurate, as well as customizable by its customers.
Custom AI will be developed on a much smaller scale than private AI, as it’s likely to be built by (and for) only one company. The scope of these tools will be narrow, with the tools designed to focus on taking care of repetitive tasks related to that company’s clients or customers. And yet, because of the narrow focus, custom AI is likely to be the most accurate of all AI types, and because of its small scale, it’s likely to be the most expensive of the three options.
Note that because large data sets are critical to the value provided by generative AI tools, custom AI is unlikely to be built by smaller firms, as they won’t have a sufficient amount of data from or about their clients or customers to provide reliable responses. (In other words, Large Language Models need to be large.)
Looking at the above structure, one could consider combining private AI and custom AI under the same “private AI” umbrella, because the two are, in fact, both private and they’re mainly separated by size and scope. But without worrying too much about semantics, I think it’s useful to separate them here as a means of showing how vastly different the scope of many AI products will be. Some will be designed for the masses, while others will be developed for increasingly smaller subsets of users.
In my mind, the metamessage of laying out this structure is that AI is going to become ubiquitous very soon, and it won’t be found in just one or two tools that we happen to use on a regular basis. In the not-too-distant future, we’ll likely use dozens – or even hundreds – of AI tools throughout our day, even as most of them won’t be labeled as such. Ultimately, AI typically will be both invisible and effortless.
VI. How can I use ChatGPT today to improve my efficiency and productivity?
We haven’t yet made it into the “invisible and effortless” era of using ChatGPT and other AI tools. And, as we’ve seen, there are some scary tales out there about how much trouble attorneys – and all professionals – can get into if they misuse ChatGPT. (If you missed it, please read Part IV, “How should I think about – and work with – ChatGPT?”, with a special close reading of #2, “You must fact-check all responses you receive from ChatGPT.”)
And yet, despite the current downsides, there are already many uses for ChatGPT that can save you and your team valuable time and money in getting your work done more efficiently.
Of course, the big question that everyone is asking themselves today is this: “How will my work, my job and my career be affected by ChatGPT?” And because law and accounting are two of the occupations frequently mentioned as being most affected by ChatGPT and other AI tools, attorneys and accountants (and their staffs) feel the weight of that question in their bones.
As to the substantive work performed by attorneys and accountants, I’ll dig into this topic in future posts, and as with everything AI-related, it will be an ongoing and ever-changing conversation as the technology becomes more and more sophisticated. For now, let’s take a quick look at the how current AI tools can help professionals with their marketing, and specifically with their blog posts.
Back in Part IV, I wrote about how important it is to proofread, polish and personalize the responses you get from ChatGPT. If you follow that advice in your blogging, you can achieve two goals: 1) Reduce the amount of time and effort it takes to write a blog post, and 2) Produce a substantive and compelling post that does more than just take up space on your blog.
For your first goal, ChatGPT (and its generative AI cousins) can help you reduce the effort you put into your blog posts. It can do this by helping you brainstorm ideas, by developing an outline of the post, and by fleshing out that outline – or, more likely, the somewhat edited outline that you want to move forward with. This can be a huge timesaver, of course.
However, it’s essential that you don’t stop there. To achieve your second goal, that of having a substantive and compelling blog post, you’re going to want to take the time to make it your own. You can do this by adding details and anecdotes that demonstrate your expertise with the subject matter and your experience helping clients with related issues. If you do this the right way, it won’t come off as boastful or egocentric, but as helpful and humanizing.
And at least for now, ChatGPT cannot quite achieve “humanizing.” So that’s up to you.
VII. What’s next from ChatGPT and generative AI?
Looking ahead, the rate of change in all things AI-related is only going to accelerate. For this reason, it’s difficult to say where we’ll be after another six months of ChatGPT and generative AI, let alone in a year or two.
But this much is clear: The appearance of ChatGPT – and the effect that it’s already had on the world – is evidence that a new era is upon us.
