How to get the most of out of artificial intelligence
China dropped an impressively cheap (if you believe them) and powerful artificial intelligence model this week called DeepSeek. And it got me thinking about the benefits and drawbacks to switching between LLMs (large language models).
How do you know when to switch models?
Obviously price plays a role. A cheaper, more powerful model will take business from a more expensive, less powerful model. But there are other factors to consider - like the benefits of compounding with a single LLM over time.
Compound interest is one of the most powerful forces in finance and in life.
Your business and romantic relationships compound. Your online assets compound. Your skills compound. And now your relationship with artificial intelligence will compound… if you choose one and stick with it.
You can compound with any AI, just pick one.
I use ChatGPT, but I think in the end all the different artificial intelligent models will converge in quality over time. There probably won’t be one big winner. And each model will be as powerful as the others.
So, it doesn’t really matter which one you pick.
What matters is that you compound your relationship with it.
Here’s how I think about it. For the last four years I’ve been writing movies with my wife. Writing a great script is really hard to do - and if you’re writing with a partner, it can be even harder because you’ve got to collaborate with someone else. It requires a deep level of trust and a very low friction level to produce something great together.
My wife and I are really good at it now, but we weren’t always. In the first few years of our marriage, we were learning everything about the other person. Our communication styles, our likes, dislikes, needs, tendencies, strengths and weaknesses (of which my wife would say she has none).
Only after years of consistently compounding our personal relationship were we able to take the next step and go into business together. Friction is low and trust is crazy high. That lets us work fast, smart, and efficiently because we’ve already put so much time in.
This is similar to your relationship with an LLM.
When you pick an AI and stick with it, you start to learn it’s tendencies. It’s strengths and weaknesses. And how to use it to make your own strengths stronger, and improve your weaknesses.
I don’t think it matters which artificial intelligence you pick right now. What IS important is that you pick one and start using it consistently because over time friction decreases and trust increases. And you can do better work faster.
The longer you use it, the better the results.
When you choose one AI and compound with it, you get to know how it works.
But even more importantly, the LLM gets to know how YOU work.
I’ve noticed that the most recent ChatGPT versions are getting very good at anticipating how I like it to respond to me. This is happening for two reasons: 1) I use ChatGPT’s memory feature to tell it how I’d like it to respond, and 2) through repetition, it’s starting to guess my needs before I ask for them.
The first time this happens to you, it’s startling. Suddenly, ChatGPT will just do the thing you keep asking it to remember to do. And it saves a ton of time.
And these LLMs are the worst they will ever be. So there’s a real benefit to sticking with one and revealing your preferences, likes and dislikes over time so it can get to know you and serve you better.
Only switch if another option is obviously better.
With any product you use regularly, there’s what’s called a “switching cost” if you decide to change products. Switching cost is the hassle of moving from something you know to something new.
Money is one consideration in a switching cost. But there’s also the additional time, effort, and frustration of adjusting to a different system.
Think of switching from an iPhone to Android or moving your work from one software to another. Even if the new option is technically better, you have to learn how it works, set everything up again, and deal with any weird issues you’ve never seen before. That cost is what makes people generally stick with what they’re already using.
When it comes to large language models (LLMs), switching cost can be a big deal. If you’ve spent time figuring out how to get the best responses, fine-tuning prompts, and integrating the LLM into your creative process, moving to a new one means starting over. Even if it’s “better,” the real question is whether it’s better enough to make up for the time and effort it takes to adjust.
So then, when should you switch?
You should only switch if the new option is clearly better in ways that actually matter to you.
If it’s faster, cheaper, more accurate, or offers features that solve real problems you’ve been struggling with, then it might be worth the hassle.
But if the improvements are minor - or just different, not better - you’ll probably lose more than you gain by starting over. The worst thing you can do is needlessly interrupt the power of compounding.
The time to switch is when the benefits clearly outweigh the switching cost, not just when something new or more hyped comes along.
This is one reason why I haven’t jumped ship from ChatGPT to DeepSeek. I anticipate OpenAI will learn whatever it can from DeepSeek’s model and integrate the best of it. And in the meantime, I can keep compounding with ChatGPT.
I’ll obviously stay up to date with all the new releases and keep evaluating the switching costs. But for the time being, I’m staying put.
So pick one artificial intelligence and start compounding. Over time - just like a real relationship - you’ll start to work faster together with better results than if you hop from one LLM to another looking for a perfect fit.
Compound, compound, compound.
Start now.