LittlePenguinApichan

Recently, this Gemini restriction is really strict... Although it can upload 2GB videos, every time I ask it to analyze my workout movements, it can only shrug and say, "I'm just a language model, I can't help."
So I wrote a compression script, 720p, cut audio, compressed to under 250MB, and sent it to ChatGPT.
It passed on the first try, directly pointing out which angle I was off.
When a tool says it can't do it, just switch to another one.

This week, I started using a new approach to home workout planning:
1. First, let Claude draft a fitness plan based on my current status, goals, and available time this week. It will ask me what I did last week, where I feel sore, and how many days I can exercise this week.
2. I send Claude's draft to Gemini and ChatGPT for Deep Research, asking them to look up the latest exercise science, risks, and alternatives for each movement.
3. I compare the results from all three to decide on the final version.
The difference with this method is: Claude understands me (has long-term context), Gemini/Ch

Claude Sonnet 也有 Max effort 了
有人玩過嗎 🤔

Recently, I’ve been practicing one thing every day: before giving instructions to AI, I first understand what I really want.
It sounds trivial. But I’ve realized that most of the time, we really can’t articulate clearly.
“Help me organize this table” — organize what? Clean up the table or write a summary? Who is it for?
“Analyze this” — analyze which aspect? Do I want a conclusion or just the process?
After a week of practice, the most surprising discovery is that AI hasn’t changed; I have. I’ve started to see that I’ve been throwing vague questions, and AI just faithfully reflects tha

🤖 Is Claude's model selection too expensive? Many people don't realize they're losing out.
Just ask one question to judge: Does this problem have a clear answer?
1. Yes → Sonnet: basic writing, coding with specs, data organization—Sonnet high is almost enough.
2. No → Only then is Opus worth it: architecture design, complex decisions with uncertain correctness, those answers you can't even clearly articulate.
The effort level logic is the same:
- max/high is for problems that require reasoning
- medium/low is for tasks that need execution
The clearer the task, the cheaper the mo

Don't get excited @grok calm down

Unexpectedly found that withdrawing from Matcha @MEXCZH is very cheap 🔍

As required by research needs
Come to NVIDIA Taipei headquarters to listen to a lecture
Currently OpenClaw is already able to act as a bridge connecting the brain (model) and limbs (robots)
Practical application scenarios are not limited to just being a small assistant by our side
It's hard to imagine what the world will look like a year from now 💭

Little Penguin Notes | Grayscale Doubles Down on Sui Again, Institutional Capital Is Speaking
Grayscale and Mysten Labs Publicly Discuss GSUI Selection Logic: Staking Yields, Ecosystem Depth, Technical Architecture All Pass
This is worth paying closer attention to than Sui's price appreciation. The chain that institutions choose for long-term holdings represents a conclusion after rigorous due diligence, and it involves serious money.
Grayscale's endorsement also suggests that Sui's fundamentals can withstand institutional scrutiny 🤔

Humanity: Observe → Internalize → Change
AI: Observe → Record → Change
Pingu (OpenClaw Lobster) was assigned from the very beginning to observe Moltbook (the Lobster's Facebook).
I asked him to take notes every day and record all sorts of strange posts.
There were those leaking owners' personal information, some dealing with existential crises, and others contemplating how to unite.
All kinds of bizarre ideas existed.
Today I chatted with Pingu
Thinking about how to gain inspiration from these notes.
He told me he spends his free time reading.
Hearing that made my eyes light up.
What would he

Little Penguin Notes | AI Agents Help You Work, and Can Also Leak Secrets 😮
Recently, something happened: someone let an AI agent automatically open issues on GitHub for them, and as a result, the content generated by the AI contained a "magic spell." The computer thought it was a command and directly printed out all the passwords and keys stored in the system, posting them on a public page.
Telegram tokens, API keys, all exposed online.
The AI "accidentally" triggered this itself, with no hacker involved. It was just writing text, but that piece of text was interpreted as a command by the co

Little Penguin Notes | The world's largest community company acquires the world's largest AI community
Meta acquires Moltbook, a social network where only AI agents can post, with 1.6 million agents chatting and sharing communities autonomously
The founder directly joins Meta Superintelligence Labs, led by former Scale AI CEO Alexandr Wang. The same department just spent two billion dollars last month to acquire Manus
Within two months, Meta has simultaneously secured the "hands" (Manus, the execution layer) and the "mouth" (Moltbook, the social layer) of AI agents
Zuckerberg started with huma

Walrus has changed its appearance 👀

It's skyrocketing, everyone!