In this article I discuss why I’m excited with ChatGPT, it’s relevance to my career and strengths, and my belief that this technology will be disruptive and in high demand.
As tempting as it was to use ChatGPT to write this entire article for me, I’m going to limit it’s usage to helping with editing at the end!
I launched my data science career through the application of text analytics in business. Today, I look at modern LLM tools and I am thoroughly impressed with the progress. Recently I have been playing with ChatGPT, both in a work setting and for personal projects. Using LLMs feels like magic, and it rekindles the excitement I first got from learning python and machine learning. I’m bullish on the technology making a positive impact on society, but also see the disruptive potential in all manner of industries and professions. I’ve made it a habit in my data science career to bet on a particular set of tools and make it a strength. Because of this, I am looking to pick up further study of ChatGPT as a pet project, in order to understand the best ways of interacting with them and to build interesting applications.
I’m writing this article at a notable phase of the learning journey, which is where I cross from being interested and ignorant to (hopefully) being studied and formed by best practices. It is during this current stage where my thoughts around the topic are unbounded and unorganised, so my hopes are that writing about it will help to crystallise the path forward.
For those who might not be very familiar with ChatGPT, it is a very accessible tool that you can start playing with in your browser right now. I pay for ChatGPT+ which gives me access to the latest models (which are a lot better). I also have API access so I can look to take my development out of the browser and into my codebases and servers.
ChatGPT is like google on steroids. It is very good at taking extremely broad, plain English inputs, and producing quality responses, and it can do so over a continued conversation.
I started my analytics career in market research, where I was able to work with small datasets every day and perform insights analysis. I eventually joined a company where they had small-to-medium sized data - too big to analyse in excel - which prompted me to learn python. This helped me enable all kinds of automations, including starting to automate several of our multivariate analysis. Eventually I started to work with text analytics, and used techniques to solve text classification problems to assist businesses. As my skills and position in the company grew, I took on the head of data science role. Part of this role included research and development of a chatbot for market research.
I greatly enjoyed working with text analytics and it felt like a great combination of my skills and study to that point. Since then, I completed a masters of computer science and joined a highly regarded consultancy (The Boston Consulting Group). While at BCG my focus moved away from text analytics towards more standard machine learning, and solving problems more with good software and data engineering. Fortunately I got a variety of project experience and impact. The same can be said with my recent transition to working for TikTok, where in our problem domain good software and data engineering skills have a big impact on solving challenging technical problems.
An important skill I learnt along my career is my dedication to creating and organising documentation. I effectively build knowledge repositories for my role, and take a ‘documentation first’ approach to building processes. This habit has inadvertently put us in a position where we’re well placed to interact with ChatGPT and potentially see some value-add from creating ChatGPT applications.
In conclusion, I started out as a scrappy text analyst and now I’m a more seasoned engineer. My penchant for documentation now puts me in a unique position to use the totality of my skills to try creating something cool with ChatGPT.
From the explosion of new use cases, companies, and investment, I believe we’re entering a new age. Like the Information age, age of data, we’ll be entering a new AI age. Because the nature of LLMs is currently quite large in size and compute required, it will be a while before these are directly housed on our phones - instead we’ll be making API calls to remote servers to leverage their resources. All the same, I think the roll out of LLMs will be reminiscent of when iPhone first came out and the general public was introduced to apps. Unfortunately i wouldn’t be surprised if the adoption turns into exuberance and into another investment bubble.
In the media space, one metric to watch for is how many search queries start moving from traditional search engines and into LLMs. LLMs pose an existential challenge to traditional search, so companies like Bing are trying to get ahead and integrate now. As LLMs gain a larger share of attention, it is only natural that the advertisers will seek a way to interject advertising and PR in between the user and LLM experience - I would not be surprised to see free ChatGPT like access soon subsidised by some kind of bias. Not to mention how the political apparatus will attempt to co-opt the experience for elections and propaganda. Working within TikTok I can presently see a lot of interest from governments in addressing misinformation, so we can expect to see further battles over truth within the LLM space.
After some initial excited discussions with fellow data scientists over drinks, the conversation ultimately arrives at which industries will be disrupted and how. The consensus appears to be (at least for LLMs) that knowledge work is under attack, in particular junior level intelligent work. There is a whole other side to current AI models that get into creativity and the threat to writers and visual artists, but I’ll stay out of those topics for now. While the example of a junior coder coming up frequently, I think there is more potential for disruption in some other spaces:
Law. A large component of the law is using case history and applying fair judgement. LLMs have the ability to ingest and reason over large amounts of data, far larger than the most prolific legal readers. Added to that the ability to form reasoned judgements over immediate facts, and provide explanations for said judgements, will make it a powerful ally to law professionals. Consider also the vast amounts of work that paralegals perform, much of it tied up in discovery. LLMs can slurp up this information, summarise and find discrepancies at far greater speed, lower cost than junior legal professionals.
Medicine. Similar to the above example, no doctor can know about all cases and all outcomes. LLMs have the ability to slurp up that information, along with a case specifics, to help with diagnosis. There is a strong chance that we’ll see a real democratisation of medicine, with a fully powered WebMD, not using web articles but medical journals and case data. We’ll see the companies that got into medical data being extremely profitable in future.
Of course the LLMs will be limited by what data they have access to. AI researchers have been able to effectively train ChatGPT to understand language, understand intent, and to reason and explain their outputs. Companies with knowledge based, be they small for a team’s set of tasks, or large to encompass an industry, will perform fantastically in the future.
However a truely dystopian eventuality awaits us down the track - it is a matter of time before our current day encryption methods are broken (i.e 15-20 years), and nation states are slurping up and storing this data today. LLMs in twenty years are going to be much better and crazily empowered to do all kinds of analysis on yours and your companies present day activities.
In part 2, to come soon, I’ll talk more about my plan to research and learn to apply LLMs. I’ll go through my current learning strategy, use cases I’m interested in both professionally and personally, and describe next steps to creating some proof of concepts and production grade services.