The AI Race Isn’t Fair For Everyone
I was participating in an AI buildathon where you have to build a system for one of the four or five tracks we were given. You can build any system that follows one of the tracks.
I chose to build an air risk manager for merchants to prevent them from fraud related to returns and replacement issues. You can use as much AI as you can, as long as your solution is there. I thought of the solution, built everything, and also thought thoroughly and connected everything. I used AI for most of it, as that was already told. Most of the code was written by AI.
Get HKJ’s stories in your inbox
Join Medium for free to get updates from this writer.
After I submitted my solution, I realized there is an issue with it. I’m a student, so I don’t have that much money to spend $20 or $30 per month on Claude or any other frontier models. Corporates, on the other hand, are going to use the best models available in the current scenario to test these my proposed solution and run the edge cases.
Here, I tested everything, including every edge case, for my project, and it really worked. It worked in most of the edge cases and also in a real-life scenario where I thought of cases that really happens in day-to-day life. Later, I realized that the frontier models the corporates are going to use are top models like Claude, Fable, and GPT Astra. These are the models they are going to use to test these out.
My system (that I built), was vibe-coded with a lower-tier model, isn’t going to survive those extreme scenarios. That’s when I thought that this AI race and the AI implementation isn’t fair for everyone.
The people who have money, those who are in big corporates, and those who have access to tokens are really far ahead of the people who are using a lower-tier model.
If you are doing everything with a lower-tier model, you have to prompt correctly to get an accurate result. As the model is being adjusted and made more efficient generation by generation, it is becoming better in every case. The required user input is decreasing, and it is getting very good on its own. Damn!
