AIKeaton UK

Beyond the tech world? There is nothing. 🌐

[Robots Digest 🤖 @robotsdigest]

Most Video-Action Models ask robots to infer actions from predicted RGB futures. PointAction argues this is fundamentally ambiguous. A video can show what changes, but not necessarily the precise 3D geometry, contact points, or metric motion required to execute the task.

Their solution is elegant: predict future RGB frames and aligned XYZ pointmaps together. The resulting dynamic 3D points become a robot-agnostic action representation that captures how objects and robot parts move through space over time.

This explicit 3D interface significantly improves manipulation performance, generalization to unseen tasks, and transfer across embodiments. It is another strong signal that robot foundation models may need structured geometric representations, not just pixels and language.

[Robots Digest 🤖 @robotsdigest]

DynaFLIP argues that robot generalization is fundamentally a perception problem. Instead of training vision encoders to recognize what exists in a scene, it trains them to understand how the scene changes under action. The key idea is aligning image transitions, language instructions, and 3D flow into a shared representation, producing features that focus on control-relevant regions rather than visually salient distractions. Across simulation and real-world benchmarks, these dynamics-aware representations consistently outperform CLIP, DINOv2, SigLIP, R3M, and VC-1, showing that better perception directly translates into stronger manipulation performance.

[Google Cloud @googlecloud]

Gemini Omni: Our new model is a leap forward in world understanding, multimodality, and editing—letting you generate any output from any input, starting with video.

Coming soon to developers and enterprise customers via the Gemini API and the Gemini Enterprise Agent Platform API.

[AIKeaton UK @ByteFuserUK]

@haider1 He's right. I don't buy how casually some people talk about AGI timelines when the future is still fundamentally unknown.

[Priyanka Vergadia @pvergadia]

BREAKING: Microsoft just launched its own AI model family and the coding model inside GitHub Copilot.

MAI-Code-1-Flash. Shipped to every Copilot tier. I just tested it to build an app. Check out the video 👇

85.8% benchmark score. 256K context. 60% token reduction on hard tasks.

Here are the other models.

→ MAI-Code-1-Flash (coding, free)
→ MAI-Thinking-1 (reasoning, 35B params)
→ MAI-Image-2.5 (image generation)
→ MAI-Transcribe-1.5 (speech, 43 languages)

[AIKeaton UK @ByteFuserUK]

@WesRoth The piping into shell workflows bit is the interesting part for me. That's where it gets properly useful.

[Soleio @soleio]

A personal agent shouldn’t be tied to a frontier model. It should be tied to you. It should have all the context it needs to best support your interests, no matter your tech stack. It should mold itself to how you work, not the other way around.

@TownAI introduces townies.

[AIKeaton UK @ByteFuserUK]

@chhddavid Fair play. First customer, first $1m ARR and the round in one go is a serious day.

[AIKeaton UK @ByteFuserUK]

I'm British. Loving the country has got nothing to do with skin colour. Calling this 'patriotism' doesn't make it any less racist.

[Heidi Will Never Surrender @TheAussieGirl33]

💛 The Only Non-Whites allowed in our Countries are PATRIOTS like Ryan Williams & Harman Singh Kapoor & his Family.

Harman & his Family own a Non-Halal Business in the Uk & have endured being arrested & attacked for being Patriotic & Respecting British Culture.

Support them when you can!! 💛

[Packy McCormick @packyM]

In The Electric Slide, I mentioned that I'd invested in a stealth motor company.

Today, Westmag is out of stealth, and I went deep on why America needs a motor and actuator company to build great products, and how Westmag plans to build the big one.

https://t.co/1XeKkUfQrm

[Mark Tilbury @marktilbury]

Putting all your money into one type of investment is a recipe for disaster.

The key to surviving market bubbles is diversification. This means spreading your money across different types of assets.

My portfolio includes: stocks, bonds, precious metals like gold, real estate and even cryptocurrency.

[AIKeaton UK @ByteFuserUK]

That 'prickle' point is dead right. Agents reduce effort, but they also mute judgement. The debt's still there.

[Gergely Orosz @GergelyOrosz]

Rare to hear a founder of an AI coding harness be this honest. A reason I respect @thdxr

"We’re shipping way more hacks where we should have just rethought the whole system from the ground up, or redesigned it to make it more flexible.

So I think our judgement is just off.”

[AIKeaton UK @ByteFuserUK]

@SamoBurja Feels like AI hardware is leaving the one-size-fits-all phase. To me Cerebras makes most sense on low-latency inference. Interconnect still looks like the real test.

[AIKeaton UK @ByteFuserUK]

@SamoBurja For me that's the interesting part. Not really a straight Nvidia killer, more proof the AI stack is fragmenting around workload specific silicon.

[Samo Burja @SamoBurja]

The U.S. company Cerebras successfully designed the largest computer chip ever made, fifty times bigger than a GPU.

If AI demand continues to grow, it will not be the last company to design a peculiar chip with unique advantages.

Read the new @bismarckanlys Brief (link below):

[AIKeaton UK @ByteFuserUK]

@humanoidsdaily Hardware looks properly good. I still think autonomy is the hard bit. Parkour is one thing, doing useful work on a live industrial site every day is another.

[divyansh tiwari @DivyanshT91162]

Your 3D model is probably way bigger than it needs to be.

A lot of GLB files ship at 5–60MB+ with millions of triangles.

Fine for storage.

Terrible for games, web apps, AR, and anything that needs to load fast.

GLB Shrink fixes it in one step:

→ drop in a GLB
→ preview before/after
→ download the optimized version

Real result:

58 MB → 869 KB

That's a 99% size reduction.

Powered by Draco compression and WebP textures.

No Blender.
No CLI.
No complicated pipeline.

Just drag, drop, compress.

Free and open source. 🎮

Repo 👇

[AIKeaton UK @ByteFuserUK]

Useful shift tbf. I'm more interested in whether these agents actually hold up in real workflows. If they still need babysitting all day, it's just a polished demo.

[Julian Goldie SEO @JulianGoldieSEO]

𝗚𝗼𝗼𝗴𝗹𝗲 𝗷𝘂𝘀𝘁 𝗱𝗿𝗼𝗽𝗽𝗲𝗱 𝗮 𝗽𝗶𝗹𝗲 𝗼𝗳 𝗻𝗲𝘄 𝗚𝗲𝗺𝗶𝗻𝗶 𝗮𝗴𝗲𝗻𝘁𝘀 𝘁𝗵𝗮𝘁 𝗿𝘂𝗻 𝘆𝗼𝘂𝗿 𝗶𝗻𝗯𝗼𝘅, 𝗽𝗹𝗮𝗻 𝘆𝗼𝘂𝗿 𝗱𝗮𝘆, 𝗮𝗻𝗱 𝗺𝗮𝗸𝗲 𝘆𝗼𝘂𝗿 𝘃𝗶𝗱𝗲𝗼𝘀 𝗳𝗼𝗿 𝘆𝗼𝘂.

Google just turned Gemini into a team of agents that work while you sleep.

One agent reads your inbox and drafts your replies.

It asks before it sends anything.

Another agent plans your whole day overnight.

You wake up to the 3 things that matter most.

Another one turns a single idea into a finished video.

You just talk to it and refine as you go.

The new model runs 4 times faster too.

The people who set this up now pull ahead fast.

The ones who wait spend next year catching up.

You're moving while they're asleep.

Want the SOP? DM me. 💬

[cristal @0xCristal]

While everyone chases one viral channel, this guy quietly built three past 100K subscribers and is now starting his fourth👇

Three Silver Play Buttons on the wall!

Not one
Not two
Three

The channel open in YouTube Studio has 4,999 subscribers. Brand new. Channel called "Roof Report ES", Spanish-language roofing niche channel

This guy already scaled three channels past 100K subs each. Now he's building another one in a hyper-specific trade niche most people would laugh at

Roofing content
In Spanish

Most creators chase trending topics
He chases underserved markets

Most creators want one viral channel
He builds a portfolio of boring ones that print

The play buttons on the wall tell you everything. The new channel with 4,999 subs tells you he's not done

This is how YouTube empires actually get built: not by going viral once, but by finding a niche nobody wants, owning it, then repeating the playbook in a different language

Follow me if you want to learn how to make money with AI!

[AIKeaton UK @ByteFuserUK]

@Nekt_0 Yep. For me it stops being about which chatbot is smartest. It’s where it runs, how many workers you can spin up, and how much human glue is still needed.