The basis of white collar work is changing rapidly
In the early 2000, with the internet and google, the grunt work around finding information was removed. Value add for any type of work shifted to putting together this information in a valuable format
For investors, this meant that bulk of your effort shifted from finding information to synthesizing it to arrive at an investment decision. The front end of the workflow – Finding annual reports, data points which used to be manual was now available at the click of button.
In the same manner, for jobs like coding, we have repositories for a lot of the boiler plate code. A significant part of such jobs is now is in glueing these components together to achieve the desired outcome
Paradigm shift
The launch of LLMs in 2022 is changing the core of all white collar jobs again. The difference this time is that it is faster and moving up the value chain at the same time
I was initially curious about these new tools and started experimenting with them in early 2023, as I did with the internet and google in the past. For those who saw the early internet, these tools felt like the dial up connection of the late 90s – slow, clunky with limited usage
Google and broadband in the early 2000s made the internet what it is today – cheap, easy to use, ubiquitous. I am seeing the same transformation in the LLMs, but at 10X the speed
The early chatgpt was Realtime and good at answering questions for which the answers already exist on the internet (and thus part of its pre-training). With the launch of the O1 and now O3/O4 models, we have reasoning models which can ‘understand’ your questions, plan the tasks and decide which tools to use to best answer these questions
This is a paradigm shift on how computers work
All other software tools follow a fixed information flow via logic embedded by the developers and system designers. In contrast these tools operate more like us, than traditional systems. They are becoming autonomous agents
Burying head in the sand
There is a lot of chatter around the implications of these tools on the future of work. I will not get into which jobs will or will not get replaced. Time will tell
A few things are, however, clear based on the current state of these tools
- The base models continue to improve rapidly based on new algorithms and more compute
- We have new reasoning models which continue to improve based on reinforcement learning techniques
- The cost of these tools continue to drop exponentially (almost 90% per year)
This means that the cost of performing routine tasks and synthesizing information is dropping rapidly. If the major part of your job is to use existing information and put it together in a different format, you face competition from these tools which can do a good enough job at 5% of the price (and dropping)
This does not mean we are doomed to irrelevance as the tools get better. However it does mean that we need to re-think what is our value add (to get paid well)
This is similar to waves of automations in the past – Farm and factory workers were not happy when machines replaced human labor. They fought this change tooth and nail. We will see the same happen with white collar work.
A lot of pushback is on the following lines
- The work quality of these tools is poor (same as weavers complaining about the quality of hand-woven cloth versus the machines)
- They are taking work away from hard working people
- It is unfair
I am not denying the pain these tools will cause in the workforce, but burying our head in the sand is not going to change reality.
Change your workflow
I personally think we should all take these new tools seriously and start learning as much as we can on how to use them. The next step is to breakdown your own workflow into what can now be done more efficiently using these tools.
Let me take investing as an example
The job of portfolio managers/Investors/Research analyst shifted from finding information to synthesizing it in the last few years. There are screening tools, financial websites, charting tools available where we can get all the necessary information in a few minutes (which used to take hours and days in the past)
The main job for us was to put synthesize all this information and arrive at the final decision – should I buy the stock, how much of it and at what price ?
As an investor, we get paid for our decision, not for the effort we put it. If we can reach a high-quality decision in a few hours versus days then it’s even better. In such a case, these new tools are a great benefit to us. We need to drop the mindset from our school days: grade = amount of homework. In markets, it is always quality over quantity
In the past I would read up a lot of documents and think of questions to answer. I would then dig further for the answers, but generate new questions at the same time. Invariably there would be a point of diminishing returns after which I would decide with 70-80% of the information
I am no longer constrained
My job as an investor is to read the necessary documents as a starting point and come up with a list of questions. I can feed these questions to one of the LLM tools and get a detailed answer. I can dig into this output, push my understanding forward and generate a new set of questions
The result is that I can have a better understanding of the company and its industry in a much shorter period of time. What can be better than that?
I will dig deeper in my next post into how I have changed my workflow and incorporated these tools.
The most important change for all of us, including investors, is now to come up with high quality questions. We are getting to the point where our computers will generate better answers than most humans