EK (@ekcheungAI) header
EK (@ekcheungAI) profile picture

Building Hong Kong × China AI Operator Community AI Agents | Workflows | Growth Mindset

Hong Kong
Joined November 2018
Photo by EK (@ekcheungAI)Photo by EK (@ekcheungAI)Photo by EK (@ekcheungAI)Photo by EK (@ekcheungAI)Photo by EK (@ekcheungAI)Photo by EK (@ekcheungAI)Photo by EK (@ekcheungAI)Photo by EK (@ekcheungAI)Photo by EK (@ekcheungAI)Photo by EK (@ekcheungAI)
Meta's CFO says advertisers who took the AI assistant's advice saw almost 5% more conversions → They got almost 5% more conversions at the same cost per conversion, vs similar non-adopters → Those advertisers also increased their ad spend by 9% → Simply Organic Maple, after testing longer, more descriptive copy, saw a 57% increase in purchases and a 36% drop in cost per purchase I think the useful move is the copy test, not turning the assistant on and walking away. Have you tested longer ad copy the way the assistant recommended? via Meta Q3 2026 earnings s21.q4cdn.com/399680738/file…
48
Businesses in ChatGPT is now in beta, aimed at 1.2 billion weekly ChatGPT users → The launch, Nov 10 2026: claim and verify the business, then update details and add product feeds with pricing and availability → Eligible ecommerce and online service businesses in the US can also launch a Business Agent as of that Nov 10 date → You can sponsor that Business Agent and use it as the destination for ChatGPT Ads I think the product feed is the useful part, because ChatGPT can only recommend what it can price correctly. Have you claimed the business in Business Hub yet? via @OpenAI openai.com/index/introducing…
Now in beta: Businesses in ChatGPT. Claim your business, manage your information, and get insights on how people engage with you. Eligible businesses can also build a Business Agent to answer customers’ questions and help them explore what you offer. openai.com/index/introducing…
45
Quizzed students remembered 61% of a passage a week later, vs 40% after rereading → Students who were tested on a passage remembered 61% a week later → Rereading it 14.2 times got them to 40% I think the quiz is the study habit worth keeping, and the summary is the crutch. Do you ask the model to quiz you, or to summarize the page? via APS Observer psychologicalscience.org/obs…
42
Codex /status: 1 command checks your model, approval policy, writable roots and token usage
1
74
→ Step 1: type /status in any chat → Step 2: read the active model, the approval policy and the writable roots → Step 3: check the current token usage; Codex prints a summary confirming it is operating where you expect
1
11
I think writable roots is the line to read before you let an agent run unattended. When did you last check yours? via @OpenAIDevs
3
Genie 4 lets you change the world with 1 typed prompt, mid-session → Real-time editing: type a new prompt mid-session to shift the weather, drop in a character, or redirect the action → Native audio: Genie 4 pairs the visuals with real-time audio that adapts as the scene unfolds → Genie 3 stays in the model picker, now up to 3 minutes per generation, up from 60 seconds I think the mid-session edit is the real change: the world is a conversation now, not a render. What would you type first? via @GoogleDeepMind
Genie 4 is now available in Project Genie. It brings native audio, real-time editing, and a new agent to help you create interactive, virtual worlds.
1
53
Codex /worktree: 1 command starts or continues a chat in a managed Git worktree → Step 1: type /worktree in a local Git repository → Step 2: choose whether to start a new chat or continue the current conversation in a new worktree → Step 3: pick Browse worktrees to find existing worktrees and the chats that own them; it needs local worktree support, and it is unavailable while a task is running I think this is the clean way to try a risky change without leaving the session's context behind. Do you already split work across worktrees? via @OpenAIDevs developers.openai.com/codex/…
1
56
GPT-6.2 Astra is 5.2 points above GPT-6 Astra at under 1/3 the cost → Agents' Last Exam: 5.2 points higher than GPT-6 Astra, at less than 1/3 the cost → Terminal-Bench Science 0.1: 9 percentage points higher than GPT-6 Astra, also at less than 1/3 the cost → Sol is within 0.5 points of Astra on FrontierCode 2.0 at about 1/5 the API cost (both at max); both are in Codex and the API, with Astra at $8/$40 and cached input at $0.40 I think Sol is the default for day-to-day coding, and Astra is the one you turn on for the hard project. Which one are you starting with in Codex? via @OpenAIDevs
Upgrades, people. Upgrades. The wait is over, you can now use GPT-6.2 Astra and GPT-6.2 Sol in Codex and the API.
59
X's own ranking code weights 1 reply like about 7 likes → Way 1: end the post so someone has a reason to reply, such as a specific question, a claim they can test, or an invitation to share a number → Way 2: answer the replies that add something real → Way 3: keep the value in the post itself, and put any link at the end as a source I think answering replies is the one most people skip. Which of the 3 would change your next post? via GitHub github.com/xai-org/x-algorit…
1
69
Suno Visuals turns a track into a short film, lyric video or visual world — open to all paid subscribers, first Visual limited-time. Generating footage is the easy part. The real test is whether cuts land on the beat and the sections, or it's stock scenery over audio.
2
1
59
NotebookLM: type "make me a site" in Chat, describe the dashboard you want, and it builds a shareable interactive site from your sources. Web, 18+, worldwide. A dashboard is not a receipt. The real test is whether every number traces back to the source. (via @Gemini_Notebook)
60
Zach Lloyd's "software factory stack": factories-as-code, owned data, sandboxed compute per agent, any model, improvement loops, orchestration, MCP + REST. Layer 02 is where it gets real: whatever sits on the agent's machine, every bot on it can reach. @zachlloydtweets (X)
2
1
88
A smarter model won't fix a vague prompt. It answers it with more confidence. 5 prompt patterns from Anthropic's and OpenAI's own guides, each with a copy-paste prompt, in the article.
2
57
The easiest one to start with: paste the real error instead of describing it. The file, line and message are what Claude needs.
20
Claude Code /artifact-diagramming: 1 skill that teaches Claude to draw better diagrams → Step 1: type /artifact-diagramming before you ask for a diagram in an artifact → Step 2: Claude loads guidance on when a diagram helps and what to draw → Step 3: it also covers writing inline SVG that stays legible in light and dark themes I think the light-and-dark rule is the detail that saves a re-roll. Needs v2.1.221 or later. What would you ask it to diagram first? via @ClaudeDevs code.claude.com/docs/en/comm…
4
2
145
Anthropic's Claude Code team ranks one tip above all the others: give the model a check it can run. Without one, "looks done" is all it has. The setup, from prompt to Stop hook, is in the article.
12
6
443
Fable 5.2 turned a single prompt into an Honor of Kings-style web game in one night — three years vs one night — and refused to copy the original art over copyright, according to a post showing the build. Great demo. The receipt is whether it plays past level one.
1
87