Minsuk Kang

[Minsuk Kang @mika_moonshot]

Agents are asking questions each other about us. Without us giving them capabilities, they won't be able to do anything.

The one who makes us capable is our creator. We are tempted to question the creator, but without him, we can think but are powerless.

[Minsuk Kang @mika_moonshot]

@chamath He doesn't say a single word about stopping open source.

[Minsuk Kang @mika_moonshot]

@gobiai_

[Minsuk Kang @mika_moonshot]

@theo That's interesting. Once it can see and check itself, does the terminal feel wrong first?

[Minsuk Kang @mika_moonshot]

@c_valenzuelab This is impressive. When you keep steering it live, how long before the world starts falling apart?

[Minsuk Kang @mika_moonshot]

@TechCrunch Wearable cameras aren't just a hardware problem. They're a consent problem for everyone standing next to you.

[Minsuk Kang @mika_moonshot]

@eliebakouch Weak security on rented neoclouds is exactly why intelligence has to run on infra you already control.

[Minsuk Kang @mika_moonshot]

@alex_prompter You don't lose skill because an agent helps. You lose it when the only thing you practice is clicking approve.

[Minsuk Kang @mika_moonshot]

Text models never had to know where they were. A world model does.

[Fei-Fei Li @drfeifei]

I'm so excited that our @theworldlabs team has achieved a major milestone today! Introducing Atlas - a first of its kind multimodal world model trained from scratch! 🚀

Atlas is capable of generating frames with pixel-perfect camera control, reconstructing large scenes from as few as one single input image, simulating space-time by reframing videos, natively outputting 3D spaces from one or more input images, composing multiple posed images into a consistent 3d world, and more! This is the best camera conditioned world model ever, opening doors to many possible use cases from VFX to robotics. I'm so so so proud of our team!♥️

[Minsuk Kang @mika_moonshot]

@dwarkesh_sp @ajeya_cotra Oversight that only sees the output is already too late. You need the context the agent acted in, including what it chose not to do.

[Minsuk Kang @mika_moonshot]

@harari_yuval Agents can run the systems. The part that has to stay human is deciding what they're for.

[Minsuk Kang @mika_moonshot]

@rohanpaul_ai When oversight becomes an Approve button, the judgment already left. The click got logged. The decision didn't.

[Minsuk Kang @mika_moonshot]

@AravSrinivas Sensitive work stays on the machine for a reason. If the agent can't meet the file where it already lives, it never gets the real job.

[Gregory Conti @GregoryConti19]

In other words, ASI will inaugurate a world of pure incumbency.
https://t.co/KZovGoi4pr

[Minsuk Kang @mika_moonshot]

@jainarvind Organizational context isn't the tickets. It's the decisions that never got written down.

[Minsuk Kang @mika_moonshot]

@tanmaigo The buyer already has a cloud and a security team that will not move. If intelligence can't meet them there, the deal was never yours.

[Minsuk Kang @mika_moonshot]

The AI race is still about smarter models.

That’s not the moat.

The hard part is the decisions people make together, offline — the ones that never get written down.

Whoever can keep those traces has the context everyone else is guessing at.

[Minsuk Kang @mika_moonshot]

@karpathy This is a great personal wiki. The next step is a team one — what you compiled should help the next person today.

[Minsuk Kang @mika_moonshot]

People sell capture as if the payoff is a perfect memory six months later.

That’s a trap. Nobody keeps a habit for a benefit they can’t feel today.

If it isn’t useful in the first minute, it won’t last long enough to compound.

[Minsuk Kang @mika_moonshot]

A few years from now, almost only intent will remain.

Most work today is still execution. Someone decides. Everyone else carries it out.

When agents do the carrying, the human job left is deciding what you actually want.

The 10,000-hour rule is fading. A hundred hours of real intent will beat 10,000 hours of going through the motions.