AI Hackathon Journey

Our design of the project evolved for 3 stages.

At first, I was inspired by Turing Test and LMArena. Turing Test provides a method to measure the ability of LLM, while LMArena proves that tech monarchs are willing to pay for the rank.

So, that's the first stage. we planned to build up a platform to evaluate models' capabilities of mimicking and acting.

And soon the second idea emerged. There exists a kind of ai -- character ai. In the past, we had to feed chatbot with a whole book and teach it the way to learn to act. But what if we train a model by the data my game collects? I mean: players judge the opponent only in some cases. And these data can be used to train an AI with human-like discriminating and discerning ability.

We went on. I built up the website totally via vibe-coding. I used a pusher, mongodb, vite + React.

We didn't advance in round 1. The prototype is plain, to be honest. And The ui, the procedure donot reflect our theme( two opposing factions). But we succeeded in round2. I credited it to zlc ---he used gemini to render our whole game.

So I think humans are visual-driven. The product should be attractive, at least, at first glance.

In second period, we have to intensively develop our product in two days. The first problem we face is the niche we want to find. First, we still hold that we can collect data or rank AI models. And even the insturctor at first agreed that we need a kind of rank aimed at small-B: those OPC need to know which model best suit their business. But it also lacked viability because complex tasks could not be evaluated only through role-play. And for training, the idea was clear: we even haven't role-played with AI. If we haven't figured out the boundary of current AI, how can I optimized the game?

And a question by the instructor inspired us: MBTI test is so long that it is almost disgusting. But why do we still want to finish? Cause we need it. What kind of results do we thrive for? We need to make players "need" it?

So we conceived an idea: make a test to evaluate how much you have been assimilated by AI: become more reactive, become more impulsive and less empathetic, etc.

In retrospect, we found some problems: this was not so interesting. And a major reason for the popularity of mbti is that they can serve as a vehicle for socializing -- you are E and I am I, we complement each other!
Besides, our game adpoted AI to assess your quality : how strange, we are judged by AI. My hindsight is that we should choose some pure mathematical ways to calculate and compare.

One pity was that I haven't explored other teams' ideas in depth. Scarely had the competition ended when I realized the most valuable thing was not win, but the ideas and opinions from innovators.

But we truly gain much. We collaborate and innovate. We clinch and pitch.
Keep going.