Stock Market Gold Miner Game 🎮

I saw everyone using AI Vibe Coding to create various desktop pets and local mini-games, and I was quite envious. So, in my spare time from investment research, I did something unorthodox and used LongbridgeAI to make a mini-game (actually testing LongbridgeAI's capabilities 🤣).

This game is called "Gold Miner: US Stock Investment Edition" 🪎

You've probably all played Gold Miner before. In each level, we have to grab the items with the highest total value within a limited time: large chunks of gold, diamonds 💎, and moles holding diamonds.

In the stock market version of Gold Miner (US Stocks), what measures value isn't gold 🏅 or diamonds 💎, but changes in company stock prices.

Moreover, each level corresponds to a different year: Level 1 corresponds to 2021, Level 2 to 2022, and so on, with Level 5 corresponding to 2025.

In each level, 10 company 'gold blocks' will randomly appear underground.

Their value is determined by the annual stock return rate for the year corresponding to that level.

Each level has a time limit of 20 seconds.

⚒️ Miners need to grab the batch of targets with the highest yield.

The yield for each level = Average (the annual returns of the targets grabbed by the miner in this round).

If you know clearly which company has the highest yield, grab only it to maximize your return.

Not sure who will rise the most? You might want to grab more company gold blocks, exchanging diversification for more stable returns.

But never grab those negative-yield 'money-losers'.

Final Game Yield = Level 1 Yield * Level 2 Yield * Level 3 Yield * Level 4 Yield * Level 5 Yield.

During this process, here are the good points of LongbridgeAI:

1. When calculating the annual stock return, it correctly handled forward-adjusted prices for targets like NVIDIA that had stock splits.
2. It successfully outputted a Demo and performed initial optimizations, generating over 500 lines of code (I initially thought this was standard, but after comparing it with two failed optimization attempts below, I realized how precious this was).

Here are the areas with poor user experience:

1. Delivering files wasn't very convenient 📁; I could only print out all the code and then download or copy it line by line, which was very slow!

•2. [The most frustrating part] It couldn't handle large-scale computational tasks. When the calculation volume was high or steps were numerous, it easily failed. Moreover, it didn't retain any intermediate thinking process, just writing "A minor issue occurred, please try again later" 😫. Watching it run step-by-step for 10 minutes, right as it was about to finish, only to suddenly fail was truly 崩溃 (crushing). Two instances stood out: First, when fetching annual stock return data for 50 companies at the beginning, it first fetched weekly K-lines, then switched to monthly K-lines, but ultimately still returned a failure after a long runtime. Only after I asked if it could fetch annual K-lines did it finally get it right. Second, when optimizing level difficulty, which mainly relies on adjusting the number and distribution of rocks. Since the game design dictates that the hook 🪝 bounces back when hitting a rock 🪨, the rocks cannot completely block the connection path between the hook and the company gold blocks. I watched it step-by-step from manually calculating the specific layout of each rock, until it finally wrote a script to automatically verify the accessibility of each company gold block. Then, snap! It showed task failure ☹️. Letting it continue, it still failed after running for over ten minutes.

In contrast, when I asked CC (I still prefer this kind of Agent that can directly help me modify things) to modify its demo, CC quickly wrote a "Layout Generator + Accessibility Verifier".