Have you ever done some home repair, and while doing so you need something you do not have? I could be a fastener, a piece of wood, a piece of metal, anything. What do you often do, you try to be creative. You look around for alternatives, you dig through an old soup container of bolts and screws. I have done this, a few months ago the bottom of a sink cabinet was ruined, and i found old shelves that i was able to cut to replace it. Instant fix. The notion that you need to find solutions with a limited universe is not new.
In the space race against Russia to land on the moon the Guidance Computers on the Apollo mission had a measly 4KB of RAM. Read that again 4KB, not Mega, not Giga but KB. The software developers had to optimize every line of code to fit in the small space. I can only imagine how hard they worked to figure it all out. They were limited by the technology that was available at the time. Sometimes limits are self-imposed, imposed by outside or just parameters you need to work within.

How about Super Mario Brothers game? The developers had to squeeze it into a 256 kilobits cartridge, and in doing so they made a lot of concessions to fit it. It added 3-4 extra weeks to development, so what if sound effects were recycled? Other examples include the first Palm Pilot (think pre-iPhone) had 128k of RAM, the first Mars rover 512k RAM, and the TI-83 graphing calculator had 32k of RAM.

Why is this the topic? In the end of December (2024) as a Christmas gift a new AI model was released called Deep Seek V3. The Chinese company that built it had was under a lot of limitations, more like restrictions. While other companies are spending billions building or leveraging massive GPU farms, Deep Seek needed to think differently based on the restrictions they were under. Without the resources of the larger companies, they found a novel way of training.

According to the Deep Seek GitHub, they only used 2.664M of H800 GPU (much older NVIDIA GPUS) for pretraining, and 0.1M GPU hours for additional training. The cost is somewhere around $5M. Compared to the model most people know, Open AI spent approximately $100M to train their current model, other models the estimate is around that same $100M with them being vocal about it costing a lot more to do the next model.
Bruce Lee said “If you always put limits on everything you do, physical or anything else, it will spread into your work and into your life. There are no limits. There are only plateaus, and you must not stay there, you must go beyond them.” I am going to disagree with Bruce this time. Limits as shown above drive creativity. Whether it is internal or external limits, those who can find the answers, exploit the loopholes think differently will break though. In every situation we are in there is a limited universe for us to work in. It is our ability to adapt, create and find solutions is what defines success. To rephrase a different Bruce Lee quote, “Use Limits as no limits”
Note: I started writing this early in January and took a step back after a few drafts and was thinking this may break a principle I keep. I did not want to promote or demean a public company. I was fortunate that others saw this, and today before this was posted the markets already reacted. My goal is for me to stop worrying about limits and find freedom in the box. If others can do it, why can’t I? I often am caught saying we cannot do this do to some limitation, I should be saying “Let me rethink this, there must be a solution within our guidelines”
The other thing I started to think about is the limits I put on others. I should think about them and ensure that there is a good reason to do so. Putting limits that someone will find the loophole is not successful, limits that are for protection, security and have some thought could be good. Thanks to all the teams who succeed given impossible limitations and giving us motivation to do the same. A friend mentioned to me that necessity is the mother of invention, but maybe the line should be necessity is the mother of creativity.
This opinion is mine, and mine only, my current or former employers have nothing to do with it. I do not write for any financial gain; I do not take advertising, and any product company listed was not done for payment. But if you do like what I write you can donate to the charity I support (with my wife who passed away in 2017) Morgan Stanley’s Children’s Hospital or donate to your favorite charity. The fundraising site had to be restarted, and NYP Hospital made changes to their donation sites. I pay to host my site out of my own pocket, my intention is to keep it free. You are welcome to comment, but note it is moderated, and all spam will be removed.
This Blog is a labor of love and was originally going to be a book. With the advent of being able to publish yourself on the web I chose this path. I will write many of these and not worry too much about grammar or spelling (I will try to come back later and fix it) but focus on content. I apologize in advance for my ADD as often topics may flip. I hope one day to turn this into a book and or a podcast, but for now it will remain a blog. AI is not used in this writing other than using the web to find information. Images without notes are created using an AI tool that allows me to reuse them.