LEAD Consulting Book Club: There’s not enough content about ChatGPT on LinkedIn
Chris Charman-Hunter
Strategic Advisor | Growing businesses through marketing consultancy
“Telling someone about your conversation with AI is like telling someone about your dreams. They don’t care; it just sounds like you’re hallucinating nonsense.” – Tom Scott
However, this week, Andy Chan selected an article where a creative strategist trials ChatGPT to help create a brand identity. Andy is what could be considered a power user of ChatGPT, interacting with it regularly to benefit his work. In this article, I’m going to summarize our conversation and hopefully provide some actionable insight into using ChatGPT effectively. At its core, the key is using it as a tool to increase efficiency, allowing us to focus on delivering more effective output.
Rise of the machine
The launch of ChatGPT in November 2022 will be looked back upon as the date on which AI went mainstream. Since then, it has amassed 100 million users in its first two months, a milestone which took TikTok nine months to achieve. This is a watershed moment, which Youtuber Tom Scott likens to Napster’s role in the digital disruption of industries in the first decade of the 21st century. The widespread usage of the technology across a variety of fields is unprecedented and marks a clear move into the mainstream.
Further to this, Microsoft’s investment in OpenAI and integration of AI into Bing shows that big tech is paying attention and making moves to stay ahead of the curve. Google’s much maligned launch of Bard also shows that the technology has a long way to go to prove its validity within a business context. However, as technology performance follows an S-curve, the future of AI application within businesses may be here faster than we think and at levels which we cannot comprehend today.
How the hell does it actually work?
Prior to the session, there were different levels of understanding of the underlying technology that powers ChatGPT. Those of us with less exposure thought of it like an advanced search engine, piecing together snippets of content from across the internet and turning it into a concise output. However, this isn’t what’s actually happening. Instead, ChatGPT breaks down the input from users into tokens and then uses the data it was trained on to predict what string of tokens gives the best answer. This is fundamentally different from how search engine results work and is more similar to the predictive text we have become accustomed to with search bars and text messaging.
A great summary of how ChatGPT works can be found here
This may seem like additional, unnecessary context, but understanding the technology, how it works, and its limitations are critical if we are to use it effectively. Our preconceived ideas coming into the session demonstrate a common fallacy when using new technology. We assume that because something is like technology we have seen before, like search engines, then it must work in a similar way. This then causes us to use new technology in a way similar to how we are using existing but fundamentally different technology. The possibilities and limitations are fundamentally different; we must change the way we engage and treat the outputs differently.
A source of information or knowledge?
Although Andy is undoubtedly the power user of ChatGPT at LEAD Consulting, we had pretty much all used it successfully in work for various purposes. Broadly speaking, there are two categories of use cases we have trialed; the first for the luddites among us was for searching for information, and the second more innovative use was searching for knowledge.
Searching for information
This is similar to the way in which we use search engines today. Looking for information from across the internet and beyond on a particular topic. In its favor, ChatGPT summarizes the results, so instead, it gives you whitepapers or case studies to provide different sources and evidence to support your views. However, ChatGPT is limited by the fact that it has been trained on data up to 2021, so the results are skewed towards historical articles. This is a challenge within marketing specifically, as it continues to evolve and change at an increasing rate. Adding to this, the results will be skewed to support the input you gave it, meaning there is a high risk of confirmation bias, where the information provided is what you want to see, rather than what you should see.
领英推荐
Searching for knowledge
This is where ChatGPT goes over and above what you can do within a search engine. Instead of looking for specific information, you can ask ChatGPT to generate new things, rather than just finding existing ones. For example, you can ask it to write code for you, such as 'write me Python code that pulls data from Google Trends.' The results aren't perfect and it needs some further instructions; however, ChatGPT comments on its code, making amends relatively easy. The possibilities of this are endless, and this is where the slew of half-baked ideas around building agency businesses that take clients' requests and automate content and campaigns sprout from. The challenge is that ChatGPT can't think critically. It is limited by what it has been trained on, and so treating it as a fountain of knowledge is perhaps premature.
Congratulations, you just got a new colleague
Where we landed in our discussion is the most effective use of ChatGPT in its current state, as mentioned in the initial article, is treating it like a junior colleague. While it may not be perfect, it can be used to do certain tasks more efficiently than a human would. We ended up with four principles for how we will begin to utilize ChatGPT and AI in general, which align with treating it like a junior colleague:
1.?????Understand its strengths and limitations – By understanding the underlying technology, we can ensure that we know what it can and can't do. A basic example of this is that ChatGPT is trained on data up to 2021, so it has limited understanding of events and development past this date. This means it's not so good at giving up-to-date information or recommendations.
2.?????The output is only as good as the input – Quality instruction is required to get the best out of ChatGPT. Giving little context or shorter instructions leads to more generic responses, as the model uses the entirety of the instruction text to predict the optimal string of words. This means you should provide as much context and nuance as possible to generate better responses.
3.?????Provide clear feedback – One of the best things about ChatGPT is its ability to take feedback and to iterate its responses. Although the initial output may not be exactly what you require, perhaps a small error in code, for example, by providing clear feedback, ChatGPT will change its output to meet your feedback.
4.?????Treat the output as a starting point – You should never just accept the output that ChatGPT gives you at face value. Some of its outputs are demonstrably wrong, and those who attempt to pass off ChatGPT content as their own are low effort and low value. However, it can be a useful tool to build from and also iterate. I even passed this article through to get some feedback, and the recommendations were solid.
Where does it go from here?
ChatGPT marks a huge leap forward in the role of AI in the mainstream. Not because of the underlying technology, but because of the widespread adoption and interest in the technology. Whether AI is a good or bad thing for our industry and society as a whole is irrelevant; it’s here to stay. The possibilities are as yet unknown and depend on where we are on the technology S-curve. I suspect we are at the very beginning of the inflection point, and the rapid adoption and use of the technology will lead us to places we struggle to comprehend.
Instead, we must focus on how we can integrate AI into our day-to-day personal and private lives to add value, rather than replace value. Whether low-effort articles and even work created solely by AI prove to be effective remains to be seen. However, for now, we’ll be using AI to be more efficient, enabling us to focus on being more effective.
Proofread by ChatGPT