Why train a model from scratch when you can merge the best of what's already out there? I'll walk you through the art and science of model merging, a powerful technique for creating highly specialized models with a fraction of the effort.
The Reality Nobody Talks About
A lot of the advice on model merging rests on assumptions that made sense five years ago. The world has moved on.
The first thing to understand is that you should focus on one thing and do it exceptionally well. I've seen this play out across dozens of companies. The pattern is unmistakable.
Many teams spend months on the wrong path before realizing that your team matters more than your technology. Once they make the switch, everything changes.
The Framework That Actually Works
Here is a simple framework for evaluating model merging. It's not complicated, but it requires discipline.
Step 1: you need to move fast and break things
This is where most people go wrong. They skip this step entirely and jump straight to execution. Don't do that.
Step 2: the best solutions are often the simplest ones
Once you have the foundation right, this becomes much easier. I've watched founders struggle with this for months when the answer was staring them in the face.
Step 3: Iterate relentlessly
Nothing works perfectly the first time. The companies that get model merging right treat it as an ongoing process, not a one-time project.
What I've Learned From 101 Companies
Across 234 angel investments, including Anthropic, a few patterns in model merging come up again and again.
The founders who handle model merging well understand that your team matters more than your technology.
What I Tell Founders
Before deciding anything about model merging, start with three questions:
What's your timeline? Because the right approach for a company with 6 months of runway is very different from one with 3 years.
What have you already tried? Most founders have tried something. Understanding what didn't work is often more valuable than knowing what might.
Who on your team owns this? If the answer is "everyone" or "no one," that's your first problem to solve.
These questions seem simple but they reveal a lot about where a company actually stands.
This connects to broader themes around DPO, model merging, small language models that I've been thinking about a lot lately.
Wrapping Up
I've shared a lot here, and I know it can feel overwhelming. But you don't need to get model merging right on day one. You just need to get started and keep improving.
Founders who do well at model merging tend to share one trait: they're relentlessly practical. They don't chase perfection. They chase progress.
That's the mindset I'd encourage you to adopt. Start where you are. Use what you have. Do what you can. And keep pushing forward.
As always, I'm rooting for you.
Frequently Asked Questions
Q:How can I apply this thinking to my own situation?
Start by identifying the core principle behind the opinion, not the specific example. Then ask yourself: does this principle apply to my context? If yes, test it in a small, low-risk way before going all in.
Q:Do all experts agree with this view?
No, and that's fine. The best ideas in business are often contrarian. I share my perspective based on my experience and data, but I encourage you to seek out opposing viewpoints and form your own conclusions.
Q:What's the most common pushback you get on this?
People often push back by citing exceptions or edge cases. And they're usually right that exceptions exist. But building a strategy around exceptions rather than patterns is a losing game for most founders.