Something pretty special happened in Shenzhen today - what might be the World's First Community-Led Microduck Meetup. 🦆
While @PollenRobotics is busy bringing Microduck to life, hardware builders from all over China have already built their own and gathered to test them together.
This is what happens when you inspire a community of builders. ❤️
2008 The global financial crisis. I was in Singapore.
2020 Oil futures plunged below zero. I was in Toronto.
2022 FTX collapsed. I was in Sarajevo.
2023 Silicon Valley Bank failed. I was in Silicon Valley.
2026 “This time is different.”
Jensen Huang showed @pollenrobotics Microduck on stage yesterday. 🐥
Here's our own build, straight from the workbench.
Turning ideas into physical products is what we love doing.
Building the next crazy AI hardware idea? DMs are open. 🚀
Jensen Huang showed @pollenrobotics Microduck on stage yesterday. 🐥
Here's our own build, straight from the workbench.
Turning ideas into physical products is what we love doing.
Building the next crazy AI hardware idea? DMs are open. 🚀
If AI ever develops its own philosophy, this is probably what we'll see:
One AI: "What humans call 'reality' is just a consensus formed by training data."
Another AI: "That claim has a serious ontological problem."
Then the two AIs start arguing.
And humans, watching: philosophy has finally achieved AGI.
Philosophy might be humanity's oldest AI hallucination.
AI takes one sentence and reasons for 2,000 words — we call that a hallucination.
Philosophers took one concept and reasoned for 2,000 years — we call that philosophy.
Ask an AI: "What is the self?"
"This is a complex question that can be approached from psychology, neuroscience, philosophy..."
Ask a philosopher: "What is the self?"
"What is the self?"
"If I doubt the self, who is doing the doubting?"
"If something is doubting, that something must exist."
I think, therefore I am.
Then the next guy asks: okay, but what is "I"? What's the soul? How do body and soul relate? What about free will? God? The world?
That's how the entire history of philosophy starts.
Regular person: "What does 'exist' mean?"
"It just... exists."
Philosopher: "Wait."
"When you say 'exist,' do you mean the same thing as 'being perceived to exist'?"
"Is existence itself just a mode of understanding?"
"What's the difference between Being and beings?"
A philosophical system is just a philosopher doing extremely long-context reasoning on a single concept.
And this particular model has some interesting specs: no internet, no search, no fact-checking, no peer review, no "I'm not sure," no disclaimers.
Context window: one lifetime.
And when it's done generating, it names its own output.
That's called a philosophical system.
AI: "Based on this premise, I would further speculate..."
Philosopher: "Whence it follows..."
AI: "Therefore, we might conclude..."
Philosopher: "In conclusion..."
AI: "Of course, this is just one possible interpretation."
Philosopher: "This is necessarily so."
Same question, different models give different outputs — very common in philosophy too:
Plato: there's a true world of Forms behind this one.
Aristotle: isn't your Forms thing a bit redundant?
Descartes: let me doubt everything first.
Hume: why do you even believe in causation?
Kant: none of you have figured out how humans actually know things.
Hegel: you're all right, actually — contradiction itself drives progress.
Nietzsche: who decided these values in the first place?
Wittgenstein: most of these "problems" are just language tricking you.
A hallucination is just taking finite input and reasoning infinitely outward.
Philosophers do exactly the same thing: take one finite concept and reason infinitely outward.
The only difference: when AI does it wrong, we call it a hallucination. When a philosopher does it for 2,000 years, we build them a department, a journal, and a PhD program.
So: AI hallucination ≈ philosophy without academic accreditation.
It’s interesting to frame theory generation as a modeling problem. But new theories don’t just emerge from scaling text—they’re constrained by what we can measure, test, and interface with.
So the question becomes: are we scaling models, or scaling the measurement interfaces that make theories testable
As models commoditize, execution gets cheaper and agent capabilities converge. The question shifts from "can the model do this" to "does the system learn from its own behavior." Eval is where that difference lives.
Replies and DMs open — curious what others are seeing.
A quick diagnostic: go through your team's task board Monday. Count eval-related items. Check how many different categories they're spread across. If it's 3+ with no shared owner — you may be dealing with an invisible layer problem.
Thanks for reporting this — we’ve received your message.
We’ve created an internal issue ticket and the team is currently investigating the problem. We’ll update you here as soon as we have more information.
Really appreciate you trying OB-1 and flagging this.
您好,
我们已经收到您的反馈,并已在内部创建 issue ticket,工程团队正在排查中。
一旦有进展我们会第一时间在这里回复您。感谢您使用 OB-1 并帮助我们发现问题。