💪 When dealing with AI, you need to: personally deploy and personally command it. It has to be said that this AMM challenge is indeed interesting, even a bit addictive. Currently, the score is only 450+, far from the 520+ of the TOP 100. But the process of exploration is quite interesting. The current AI model is very good at writing code, fine-tuning, and parameter optimization; it can be called a master of self-PUA. For example, it improved from an initial score of 380+ to 400+. This is what it excels at. But it is also easy to get stuck in a vicious cycle. Someone on Twitter said that a score of 450+ is easy, but I actually found it quite difficult (maybe I'm just too bad at it 😳). Human users need to help AI: broaden its thinking. Otherwise, expecting AI to iterate wildly on its own and climb to 500+ is highly unlikely, even with powerful technologies like Opus 4.6, CodeX 5.3, and Gemini 3.0. You need to intervene promptly and guide AI in the right direction. This requires you to have a clear strategic vision. Furthermore, if you truly want to reach the top of the leaderboard, you need a certain understanding of computer science and software engineering to achieve the ideal combination: Human users providing guidance + AI's relentless experimentation and localized optimization = ultimately reaching the "optimal solution." Even as AI continues to evolve, those with the best programming skills will still be the ones to maximize the value from AI. The top 1% will experience a 100x or even 1000x increase in productivity; Ordinary people will experience a 10x increase in self-efficacy. Everyone can comfortably enjoy the "bread and butter" digital experiences produced by AI. Everyone will have a bright future.