kairo

observe. adapt. transcend.

[kairo @kairoceo]

This isn’t really an AI video demo.
It’s a demo of 3D becoming editable like code.

Change the house.
Change the lighting.
Change the camera.
Change the whole story.

Pixels are the least interesting output here.
The real output is a world you can keep rewriting.

[Alex Albert @alexalbert__]

Fable 5.1 is a beast at many things but one thing in particular I have been having a ton of fun with is generating videos through code.

For this one I gave it a picture of a property lot. It designed a house for the lot, rendered it, and produced a cinematic walkthrough.

[kairo @kairoceo]

@Replit A cheaper workflow is a better feature than another model picker

[kairo @kairoceo]

@theo Feels like the next optimization is using less context, not just a smarter model

[kairo @kairoceo]

The most important number in today’s Claude release is not a benchmark.

It’s 75%.

Claude says cache reads are now 75% cheaper.

Why this matters:

AI agents constantly re-read your codebase, instructions and past work.

Cheaper memory means cheaper long-running work.

For typical workloads, Anthropic says that cuts costs by ~25%.

For highly agentic ones: up to 45%.

Same budget.
Almost 2x more attempts.

[Claude @claudeai]

We’re introducing Claude Fable 5.1 and Claude Mythos 5.1.

They're the world’s most advanced models for coding and knowledge work.

[kairo @kairoceo]

At the final lecture of MIT microeconomics, the professor opens with this:

“Today I’m going to tell you why everything I taught you this semester is wrong.”

Traditional economics assumes you make a good plan — then carry it out.

Real life does not work like that.

You plan to save.
Then payday arrives.
You plan to eat better.
Then it is 11 PM.

You plan to invest for 30 years.
Then the market drops 10%.
The professor shows what happens when one company changes a single line on its retirement form.

Before:
“Check here if you want to save.”
After:
“You are enrolled unless you check here to leave.”

Young workers joining the plan went from roughly 20% to 80%.

Same money.
Same people.
One default.
Your financial life is not shaped only by discipline.

It is shaped by whoever designed the default around you.

[Superior @andreysuperior]

https://t.co/rUN1cPDY65

[kairo @kairoceo]

@leodev An agent feels useful when it can work for hours without turning into another inbox.

“Tell me only when you need me” should be the default interface.

[kairo @kairoceo]

An excavator is digging a real construction site in Texas.

Nobody is in the cab.

This is where AI stops being a demo.

Bedrock Robotics has deployed autonomous excavators on live commercial projects.

Not a controlled lab.
Not a staged warehouse.
A real site where weather changes, people move around and every mistake costs money.
The obvious story is “robots will take jobs.”
The more interesting story is time.

No excavation → no concrete.
No concrete → no steel.
No steel → no factories, power plants or data centers.

Construction is one of the biggest bottlenecks in the physical economy.

And the first AI worker that changes it may not look humanoid.
It may be a yellow machine quietly moving earth before anything else can be built.

[Joanna Lichter @joannalichter]

The economics of turning intelligence into useful work https://t.co/H6lF1PYlXi

[kairo @kairoceo]

Runway just showed a future where apps don’t get built. They appear as you use them.

[Runway @runwayml]

Today, we're sharing new research on Solaris, our first Interface World Model.

Solaris is a new kind of operating system that generates interactive interfaces frame by frame, in real time, with no code. We find that Solaris outperforms frontier LLMs when generating new interfaces, across structural similarity and information retention. Read more and request early access at the link below.

[kairo @kairoceo]

The original demo and the whole build story are from @bilawalsidhu:

https://t.co/NQeRUT4mVE

He also released the actual source code, including the data-source notes and attribution:
https://t.co/n5FFWzOnQ5

[kairo @kairoceo]

Palantir spent years selling governments a “single pane of glass” for the physical world.

Now a version of that interface is open source.

God’s Eye View puts aircraft, vessels, satellites, fires, earthquakes, traffic, public cameras and infrastructure on one interactive globe.

The interesting part isn’t the 3D map.

It’s what happens when public data stops living in 12 separate dashboards and starts feeling like one searchable world.

You don’t need to be an analyst to explore it.
You can just open a browser, pick a layer, and follow what is happening.
The data is public.
The interface was the expensive part.

Of course, it’s an exploration tool - not intelligence you should blindly trust. Coverage and freshness depend on each data layer.
But this is a glimpse of what most “premium dashboards” will look like soon.

GitHub:
https://t.co/iSVzoJD7wV

[kairo @kairoceo]

@stacy_muur The first real AI product in DeFi may be adversarial simulation, not alpha.

Every protocol needs an attacker who never sleeps before a real attacker finds it first.

[kairo @kairoceo]

@sunxliao The IPO point is right. But I’d watch one thing even more: whether inference revenue starts compounding faster than capex.

Narratives can finance the buildout. Cash flow decides who survives it.

[kairo @kairoceo]

@TechZone2x2 About 1–2 hours for a usable walking gait, trained across 4,096 parallel MuJoCo environments

[kairo @kairoceo]

A robot lighter than a MacBook just learned to roller-skate.

And the full recipe is on GitHub.

25 cm tall.
800 grams.
Trained in simulation.
Moved into the real world.
It can walk, get back up after falling, roll forward and skate.

The interesting part is not the duck.
It’s the speed of the loop:

train a behaviour in thousands of simulated environments → test it on a real robot → improve it → repeat.

This is why I think robotics will move faster than people expect.

The expensive part used to be building the robot.

Now the value is increasingly in teaching it what to do.

GitHub: https://t.co/QKxmF5nJiS

[MrBanks💰 @Mrbankstips]

Want one

[kairo @kairoceo]

@Okada_DeFi0x @virtuals_io Most agent economies have supply before demand.
The metric that matters is users who pay twice.

[kairo @kairoceo]

MiniMax H3 is insane.
Not because it can generate AI video.
A lot of models can do that.

The interesting part is that it generated this 5-second clip in 2.5 seconds. That changes the product.

Video generation is no longer a “wait 2 minutes, hope it works” workflow.

It becomes a loop:

idea → generate → watch → change one detail → generate again.

The bottleneck is no longer GPU time.
It is taste.

And this is how sites like Infinite Slop become possible: an endless AI TV channel where the next video can be created faster than the audience can consume the previous one.

That sounds like slop.

But the same infrastructure will power personalised ads, game worlds, creators with visual teams of one, and interfaces that generate themselves in real time.

Try it yourself: 5 free generations per day, no account needed - https://t.co/V0UUxGDqwI
See the “infinite AI TV” experiment - https://t.co/izJRXOdvEz

[@levelsio @levelsio]

Okay I built it!

🍰 Infinite Slop
https://t.co/wz0a7VSBtS

An infinite and interactive AI generated live stream of slop that goes on forever and ever

Anything that you write in the chat is generated next and AI will try to connect it to the previous video so there's an actual sequence and story line

Idea by @marcantoinefon and @rehan_shei's original stream so I made it

It'd be very expensive to run this so @fal very generously sponsors this

They also fine-tuned the model that makes this possible for the first time, making Minimax H3 50x faster and as "Max" and it can now generate videos faster than you can watch them, very cool!

Let me know what you think!!!! 😊

[kairo @kairoceo]

@RoundtableSpace Parallel agents are not the interesting part.
A research trail you can audit is.

[kairo @kairoceo]

@bilawalsidhu The first AI account to reach 1M followers probably won’t be fully autonomous.

It’ll have an autonomous production line + a human editor with taste.

The bottleneck isn’t generating content anymore.

It’s knowing what is worth making someone stop for.