I am a subject matter expert in various topics. I don’t want to oversell. That makes it seem like the number is many. It’s not. Just the topics are quite unrelated.
This isn’t a title I anointed myself with. This title I was give by regulatory and oversight organizations. When you are a subject matter expert you often get sent coursework or textbooks and then get paid a fairly ridiculous amount of money to sit on a committee and discuss those materials before the are certified for use in the classroom or wherever they are to be used.
Everyone is a subject matter expert at something.
I am telling you this because what I am not is a subject matter expert on Python or coding. I know the difference.
Claude (and my in-house agent Fluffy) make me at least 20x more efficient at what I do. Most of what I do is data analysis. I have a thought in my head and then I try and find data to support or refute that thought.
My former process was code on Saturday/Sunday. Spend a few hours over the next two to four weeks at night debugging and iterating over that code until it actually produces the data I need.
That whole process takes a few hours now and is done in the background while I watch YouTube videos on Octopi farming. I have my weekends back.
What is more impressive is questions I wasn’t smart enough or couldn’t code well enough to answer I can now.
But, if you aren’t an expert at the subject you are using an LLM for you will get the wrong answer - almost all the time.
Earlier today, I received this message.
“You were right, and my earlier analysis was badly wrong.”
Now my guess is if I had showed the original result the agents returned maybe one person I know would have caught the error. And if anyone had acted on that result - they likely would have lost money.
The question that follows is has me using these LLMs made more money this year that last? They have made me more efficient but has that efficiency turned into more money?
And that answer is no. Or nope if you prefer.
This year’s returns after 8 months is exactly inline with the average returns over the last decade after 8 months.
This is not surprising. And I am pretty sure everyone using these tools for anything other than coding has reached the same conclusion. They make you more efficient… but that efficiency doesn’t really translate into a new set of wheels for your road bike.
The reality is no one really needs to be 20x more efficient. The physical world still exists. So you just get to wait around more.
Some of you know we used to run semi-high frequency trading bots and those bots also under performed in comparison to just me trading. So this isn’t one test case result here. I’ve been studying this for 6 years now.
What is going to happen is that the lowest 10% of workers and the most redundant of management (even to executive level), the people who aren’t subject matter experts and who can be replaced with a potted plant without sacrificing revenue will be replaced by AI. Wall Street’s head count→growth story is gone.
This is very similar to what happened with the price of Potato chips after COVID. Frito Lay decided they could raise prices because they could blame the supply chain. CEOs are going to lay people off because they can blame AI.
There is no negative brand effects if everyone is doing it.
But, in order to do this, companies are going to have provide AI to each other employee. Because even though the accuracy/efficiency of replacing those workers is immaterial the volume of work will decrease so you still have to subsidize (i.e. over work) your remaining staff.
You can’t fire people without having the excuse on the balance sheet to back that decision up.
This is where an economist would get all twitchy.
If 10% of white collar workers can be displaced by AI the US will save $1T every year in costs. This means if the AI costs of the remaining workers is $1000/month the net savings per year will be ~$250B.
This equates to a +2% rise in unemployment or ~ 6.2% assuming those white collar workers stay unemployed and down slide down the economic ladder. If this were to happen the Fed would lower interest rates allowing companies to refinance debt.
This scenario would put the total revenue to LLM companies at around $500B/year which is sufficient to stave off the bubble collapsing and buys enough time for LLMs to become more efficient i.e. cheaper.
The point is there is a possible path forward were this does not end in another Big Short movie but also doesn’t provide growth in revenue either but falling interest rates cover up that fact.
I will let you decide if those numbers make sense or not. And we still have oil and the bond market’s opinions to consider.
—AJ