The LLM Options Analogy

I recently published Short Where She Lands, Long Where She Ain’t

Why did it take so long to get around to a post I thought was important?

I knew I wanted a way to demonstrate the principle via simulation not just words. Coding is a slog for me so there was enough friction to just keep punting this. But GPT-4 changed the game. It was my assistant the entire way. Every single line of the graphing code was written by GPT. All the debugging was done by GPT. I’d give it all the code and the error and it would just give me the exact code to fix it. I’m almost shaken by how helpful it was.

The experience has given me some new ideas for things I can actually prototype that I didn’t think was possible to do on my own (I’ll of course share them when they exist).

Agustin Lebron tweeted this last week:

The LLM as a put option on cognitive tasks. Almost a year into the world’s experience with ChatGPT, it’s pretty clear that its biggest value is helping people with things they’re not good at.

It’s not going to write the next great American novel, but it will write that annoying blog post you need to churn out. – If you’re not great at data analysis, Code Interpreter will do a fine job. – It won’t make you an edge-level researcher, but it can help you learn faster.

It will write decent code for problems that have been solved a bunch of times before (React frontend, Flask backend, etc) but won’t design the perfect task-specific data structure or type hierarchy for a new problem. Basically, it makes you decent at what you’re not good at.

That’s basically the definition of a put option. It gives you a cap on the amount you’re going to lose relative to a competitive world in situations where you’re not that good. Using an LLM is like cheaply buying puts on the holes in your cognitive skillset.

So how should you rationally adjust your behavior given you own these cheap puts? Embrace variance! Your value to the world less dependent on what you’re worst at, but more dependent on what you’re good at.

So go out and hyperfocus! Whatever it is that interests you so much you focus on it to the exclusion of all else, go do it and feel no guilt. LLMs got your back.

I replied in agreement:

If you subscribe to “lean into your strengths and just spend the minimum to get your weaknesses serviceable” the LLMs just raised the strike for the same given cost.

I’ll round this out with one more idea that spares you the option analogies:

The Brilliant Math Coach Teaching America’s Kids to Outsmart AI (WSJ)

If you have anxiety about the overwhelming technological pace change this post can help reconcile your thinking about these tools and how to use them to complement what humans are best at.

Po-Shen Loh, a professor at Carnegie Mellon University and Team USA’s coach for the International Mathematical Olympiad, is traveling to 65 cities and giving 124 lectures before the next school year like he’s on a personal mission to meet every single American math geek. 

His simple advice for an uncertain future: Be more human.

Check out the post to see what he means.

I wouldn’t sit on your generative urges. It’s never been easier to take em off-leash.

Leave a Reply