02: Those AI Terms You Don't Understand, I'll Translate into Plain Language
Starting with this article, we will officially begin learning AI and build your own AI knowledge system with me.
✨ Latest
How to Choose Among These 65 GPT 5.6 Models
How to Choose Among These 65 GPT 5.6 Models
- GPT 5.6 Sol is the flagship model, representing OpenAI's strongest intelligence (What? You say there's 5.6 Pro? Pro is not a model, so Sol is the strongest.)
- Sun > Earth > Moon, corresponding to Sol > Terra > Luna (Officially positioned as Flagship > Daily > Cheap). For coding, only consider the first two; don't use the cheapest Luna.
- 'max' indicates the highest thinking effort supported by the model (before 5.6, the highest was xhigh), so gpt-5.6-sol max is the strongest full-state model.
- 'ultra' is the multi-agent mode supported by the codex tool. Don't be fooled. Enabling 'ultra' defaults to 4 agents working in parallel, and currently does not support configuring the number. AI suggests...
64 AI-generated image assets, what to do if the results are not good
64 AI-generated image assets, what to do if the results are not good First, let's look at the results. These are the images generated for me before optimization, and from another app. What can I say about this style? I'm speechless. Now look at the optimized results. See, see? The quality instantly improved.
As usual, let's go straight to the prompt. The icons and listing assets generated by pub skill are not good-looking. In this regard, find some popular, effective, stylized professional image generation prompt templates from the community for reference and adaptation. Then verify the actual effects on the target project, evaluate the image generation results of multiple solutions, and select the best performing one as the image generation solution for pub skill. Note, first conduct thorough research, then compare the effects. Only use one target project as the testing scenario: /Users/pocket/Documents/project/freeup and retain the generated images from the process...
63 How to Make Your Skill More Powerful with More Use
63 How to Make Your Skill More Powerful with More Use I previously wrote an article titled '55 How to Truly Evolve Your Skill: A Step-by-Step Tutorial.' Today I want to talk about a simpler case study, so I'm dedicating a separate article to it. Background: As everyone knows, I have a copy Skill that is used to replicate any website with one click. https://ps.useai.online/ https://maliang.useai.online/ https://app.useai.online/ These are all masterpieces created by the copy Skill. Problem: My copy version occasionally experiences a loading bar blocking the canvas, making it inoperable. So, starting from this problem, after solving it, we can then feed back into the copy skill. Let's first look at the effect. I glanced at the rules, they were what I wanted, and the key sentence is: Fo...
62 Create a UI/UX Skill, Step-by-Step Tutorial
62 Create a UI/UX Skill, Step-by-Step Tutorial There are probably no professional designers here; coincidentally, I'm not one either. But I need AI to design good-looking UI effects and bring the product's interactive experience to 70-80 points. So what should I do then? This article will discuss this typical scenario: how to piece together a Skill that fits your own use case. First, I have some input from group members, who mentioned two very powerful UI skills. One is "Impeccable" from a Google Design Director https://github.com/pbakaus/impeccable The other is the highly-starred pro max https://github.com/nextlevelbuilder/ui-ux-pro-max-skill...
61 I don't know algorithms, but it's okay, my AI does.
61 I don't know algorithms, but it's okay, my AI does. As everyone knows, as a frontend developer, I have no background in algorithms, and I've always admired algorithm experts because I don't understand Bayesian, Fourier, annealing, neural networks, feedforward, feature classification, deep learning... just by my incoherent speech, you can tell I'm truly an algorithm novice. A beginner wanting to play around wouldn't have worked in the past. But this time, I've waited for AI. I might not know algorithms, but can't I still state my requirements? Let's do it. In the previous article, '59 Untold Secrets: How to Tame an Agent', I mentioned infinite nesting. At the end, I also updated a few methodological examples, which provide a stronger experience, so you can take another look. This article is also a case study, one of three things that made me stronger: a real case of (cross-industry) using AI to do things I couldn't do at all: 3. I became stronger:
- Using AI to write 100% of the code → (While I'm at work) letting AI write code at home by itself.
- Manually transferring information gaps to write public...
If you want to transform into an agent engineer, Codex is it worth learning first?
Codex is not a universal tool, but it is a key starting point for understanding Agent engineering. This article uses a specific exercise to take you through the true capability boundaries of Codex and points out the pitfalls that are most likely to be stepped on during transformation.
How does Agent Engineering work in a real AI encoding workflow?
Agent Engineering Connect large language models to safe, observable, rollback engineering systems. This article uses a real repair task link to break down its responsibilities, handover points, failure scenarios, and what you should do as the first step.
How to use Context Engineering into AI coding workflow: a set of executable implementation paths
Understanding the concept is only the first step, connecting Context Engineering into the real AI coding workflow is the key. This article starts from the point of failure and gives the minimum landing path, the actions most likely to fail, and backup plans, so that you can get started directly.
How does Web Search Tool For AI Agents work? I'll make it clear with a real link
The Web Search Tool is a key component for AI Agent to connect to real-time information. This article uses a real execution link to dismantle its working principle, common failure scenarios and specific avoidance methods, and gives the first step to implement it from scratch.