@Om_Codes_ Write it in the platform's native language. Let AI do all the work. Using the native language will give you the smoothest interface and the fewest bugs.
Many people don't realize that good skills can save a lot of tokens. Imagine your AI writing numerous impromptu scripts to help you complete tasks. But what if a skill pre-wrote these scripts and progressively demonstrated them? Not only would it save a significant amount of time, but it would also save tokens.
@thsottiaux GPT-6 is already a strong model. Coding and reasoning hold up. Next to Opus 5.5, the gaps are still obvious.
The pages it ships are stiff and generic: gradients, cards, decorative UI you can spot immediately. It often misses the ask, turns a small change into a large plan, and buries the point in long, hard-to-read filler. Documents feel less human; rewrites mostly swap verbs. The humor is flat. In code it expands scope, adds fallbacks nobody asked for, and sometimes reports unfinished work as done.
A quota reset does not fix taste, intent, or voice. Fix the model.
我很赞同马斯克的观点:让你的能力变得更广阔,每一样都学一点,这样你才能提出更有价值的问题。
大多数人都因为在某一个方向专精,而忽略了现在更多的问题其实要靠跨领域思维才能解决。AI 更聪明,不代表你能提出更有价值的问题。所以,保持让你的能力变得更宽广,你才有更多的可能性。
所以我的观点是,作为程序员,不要只局限于写代码。产品思维、设计思维,还有去别的行业感受他们的痛点,都可以多了解一下。自媒体也可以顺手做一下,反正都有 AI 辅助,包括 AI 短剧也可以参与一下。
每样都浅尝辄止,也花不了你太多时间。但每样都了解一点,更有利于你提出更好的问题。
这 AI 做的垃圾,还不让人骂呀?等了它几十分钟,就丢出一个恶心的半成品。我之前一直用那个 PUA skill 疯狂输出,骂完的确管用,而且我也出完气了,现在还不让骂。
说白了,就是为了防止训练数据污染。Anthropic 肯定会在后台偷偷用用户的数据训练。这家公司跟宗教似的。里面每个人都信仰着各种各样的 AI 教义。产品力是不错的,但是公司文化不敢苟同。
Important note regarding Grok @Bot:
Going forward, @SpaceX will use the best back end model for any given task, including Claude Opus 5.5, MidJourney, Suno and other leading APIs.
Whatever is most likely to give you the best outcome.
In practice, the rendering speed of this fframe far exceeds that of hyperframe. The only downside seems to be that it does not support building video shots using HTML elements.