It's time for versioning in AEC to evolve again. Models are huge and for most of us "storing a version" still means saving another file.
Object-based versioning like Speckle was the first step away from that. Split the model into objects and only save what's new so storage drops a lot.
What if we went a step further and versioned every component on its own? Components sit inside other components and they refer to and rely on each other. Blocks and families already update every instance when you change the definition. Now imagine each component keeping its own history, scripts included, and pointing at the exact version of the other components it relies on.
That's when designers, teammates and AI agents can actually work together like programmers do with Git.
New finetuning meta just dropped
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…
Jovin retweeted
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…
AEC community on X is so sad, the industry needs to be on here
Natural exposure to the possibilities and more understanding of what’s just a chat wrapper
Whenever my 5hr claude usage is up, my brain goes through the dilemma of upgrading my claude plan or have free time to zone out.
Browser based 3d apps about to takeover the internet
Today's #ThreeJS PR: High accuracy, spatio-temporal de-noised 1 ray/pixel Screen Space Reflections (SSR) re-write @ 60 FPS + coverage output.
github.com/mrdoob/three.js/p…
getcadcraft.com/
The most expensive suite of tools in the AEC industry being re-made open sourced.
They gatekept smaller firms but with this progress, likely < 3 years before they catch up and all peeps get the useful features
Jovin retweeted
i own a relatively new (2025) samsung tv.
i kept getting weirdly targeted ads.
i tried reversing the firmware but it was aes encrypted and no keys unless tear it apart and do uart/jtag magic.
zero days it is. root shell on new samsung TVs!
discovered PRIVACY HORROR. write up?
Jovin retweeted
Spent a bunch of tokens over the last few weeks building a fully procedural New York simulation.
mrdoob.github.io/slop-newyor…
It's only 784KB gzipped and the code is on github, so you can do whatever you want with it ✌️
Jovin retweeted
Everyone who spends more than 7 hours of their day on a mobile device or computer is at risk for becoming an NPC and there are absolutely no exceptions.
You have to break the pattern, do something different, do something new or else you’re cooked.
Eventually you will lose your mind, your individuality, your critical thinking skills, most of your social skills, and your soul.
Five nights at Freddies
Introducing Hugo: the first human companion robot.
At Boston Dynamics, I was Chief Strategy Officer, bringing robots into the real world.
My co-founder, Farbod Farshidian, led research at the RAI Institute.
Together, 15+ years teaching robots to run, jump, and work.
So we asked: what if robots could connect with people?
Hugo runs on the first Vision-Language-Expression (VLE) model.
He links what he sees and hears to movement, expression, and gesture.
He notices you, reads the room, and reacts in real time.
And he's only the first.
Dynamic Creatures is @BostonDynamics' official entertainment and hospitality partner.
We're building a cast of characters with film studios, resorts, and theme parks.
dynamiccreatures.com
Jovin retweeted
To those asking what SG Govt can do better
Haze started early September
1. 10 Oct - we publicly offered Indonesia firefighting help.
ONE month in
2. We set up a task force w 28 agencies, but nothing has been done much. PSI data just on news
3. Kids still do outdoor activities until PSI 200 🤔 including PE i believe
3.5 If you actually cared about the kids' lungs you'd want them in AC classrooms for all ages + stop PE the moment it's unhealthy, not 200
This is relatively annual + besides heat lowers IQ lmeow.
Haze defo affects their brain and health development.
We should’ve installed years ago since climate temps been rising.
4. Free N95 masks only kick in past PSI 200 ("very unhealthy") - and only for selected vulnerable groups
5. Schools only "considered" for closure if PSI is forecast >300
6. The 24h PSI is an average - what you breathe hour to hour can be way worse. Check the 1hr PM2.5
7. We've had a haze prosecution act since 2014. 11 years, ZERO prosecutions.
You telling me nobody can be charged?
Just cos we don’t have cooperation, there’s no creative ways to find the culprits?
8. Maybe we just dont arrest people but like go offer incentives / grants at the end of each year of ur areas has no fires and they don’t slash & burn but use other methods?
9. No cloud seeding bc we're too small and the rain might not even land on us and may fly to nearby countries. This was stated in SG parliament bef ok. bless our neighbors w water not bad what.
Just FYI dirty air is linked to lower cognitive scores and long-term heart + lung issues.
We can do way better or at least try other than pushing PSI numbers.
To be honest, existing AI agents already have access to Rhino and every command inside it. New chat agents and wrapping BIM in a chat adds zero new capability.
What these existing agents need are tools to understand spatial states, access deterministic solvers and data history.
Jovin retweeted
hear me out:
instead of democracy, we fine-tune the best open-weights model with Lee Kuan Yew’s writings
then let it run a small public sector gov
Design in architecture is rife with tight standards and frameworks. That isn’t wrong, it’s how something gets built. What’s wrong is limiting creativity and freedom of design with unmovable data structures which govern what’s even possible with design applications.
What’s needed isn’t one more software for every specialization, or another software to translate between more and more applications.