From the reporting here allegedly OpenAI has trained Astra to do latent space reasoning, which is thinking in a more abstract way and not necessarily jotting down in text all the model's thoughts (as how all the current frontier models work right now). Which is most likely a lot more information and allows the model to reason about things that can't be very well described as text. As for example when us humans do spatial reasoning we don't really think in text.

Maybe I'm getting old, but AI has probably been more useful to me fixing stuff around the house than it has been writing emails or any of the things people keep talking about. The other day I had a door hinge pulling out of the frame. I've always used the old toothpick trick because that's what my dad showed me years ago. AI suggested using gel super glue as well. Never crossed my mind. Took five minutes and it's probably the best that hinge has ever been. Same thing with cars. Same thing with plumbing. Half the time I don't actually need AI to tell me what to do, I just need someone or something to point me in roughly the right direction so I stop overthinking it. Now before anyone says I'm replacing YouTube with ChatGPT, no. AI gets plenty wrong. It once wanted me to spend half an afternoon repairing a cheap kitchen appliance that costs less than a decent takeaway to replace. It has absolutely no concept of when something isn't worth fixing.

That's probably the bit people miss. AI isn't replacing experience. It's replacing that feeling of staring at something broken and thinking, "I've got absolutely no idea where to even start." Twenty years ago I'd have been digging through random forums hoping someone had the same problem. Today I ask AI, sanity-check the answer, and get on with it. Curious if anyone else has found this. Has AI actually changed how you approach DIY or fixing things, or am I just becoming the bloke who asks a chatbot what I used to ask my neighbour over the fence?

3-axis test successful. Planning to use it to make a 6-axis desktop robotic arm...See if we can implement remote operation or simulation.

Vals AI has claimed that fable 5.1 helped solve a three centuries old cipher that no one else could figure out.

Here’s the X thread that sums everything up:

https://x.com/ValsAI/status/2094851406931095593

For those who don’t like X or want something more in-depth, here is their blog:

https://www.vals.ai/blogs/fable-solves-cyphral-distich

DIsclaimer: Fable was the orchestrator, and called itself on some tasks, but called Opus 5.0 on most of them. This still took 36% of my fable budget on a Claude max 20x sub, so doing the whole project with Fable would probably have spent all of my Fable budget or more.

I am also not certain that i have truly picked the absolute best prompt to make an impressive 3D scene but i wanted to actually make a game from it.

This was done in 2 shots. It made a first pass, i reviewed it, then it improved it again. I probably could go further.

My actual prompt is long and not worth sharing here because it reference past projects, but the main thing worth knowing is it spawned a lot of sub agents and did so in 3 waves. If someone really wants to see it: https://pastebin.com/iPnk4PZ8

This took around 5 hours and 30% of my weekly budget...

The setup: Secondhand UFactory xArm 7. A size-4 round watercolor brush in a 3D-printed compliant side mount (leans roughly 16° from the paper). A cheap 24-well tray for paint, two water wells and a sponge for cleaning, 300×300 mm paper held down by magnets. Overhead camera with a ChArUco solve for the paper plane; brush-tip height from a touchscreen contact station. Motion is planned and executed by an AI coding agent driving Python stack (path planning, FK preflight against the current tool, safety gates, logging).

Operating a brush is tricky. It's is a soft, tilted, wet tuft, and the mark depends on shaft tilt, rotation about the shaft, depth, and how much pigment and water is loaded. None of that is observable from the controller.

This particular run had 16 paint pickups, 96 palette circles (six per load), 8,437 waypoints, about 32 m of arm travel for 28.8 cm of painted line, 31 minutes wall-clock.

Atlas: A World Model for Spatial Intelligence | World Labs

AI Summary:

The blog introduces Atlas, World Labs’ new spatial world model. In brief:

• Atlas takes text, images, video-like image sequences, camera poses, and depth and builds a persistent spatial understanding of a scene.
• It can generate unseen viewpoints and infer missing geometry, rather than only reconstructing what was directly observed.
• It can turn those generated/reconstructed scenes into practical 3D representations such as Gaussian splats for fast rendering.
• World Labs emphasizes that Atlas can be updated with more observations, so its guesses about unseen areas can be replaced by real data.
• They position it for things like robotics, simulation, 3D content creation, and spatial AI.
• The key claim is that Atlas is not merely making pretty 3D reconstructions; it is learning a model of how a scene is arranged in space and using that to predict new observations.

The main caveat is that the blog demonstrates strong spatial modeling much more clearly than it demonstrates a fully general physics-based world simulator.

Further post trained on Coding & Cowork, Qwen3.8-Max-0902 now delivers stronger performance across complex enterprise tasks, scientific research, and long horizon workflows.

https://x.com/Alibaba_Qwen/status/2094968708288680276

Astra's COT is Supposedly Abstracted or uninterpretable using looped transformers, im guessing for the sake of Efficiency.
What are your opinions? Is Singularity but also the Risks becoming closer and closer?

Estamos muy cerca de comenzar las pruebas de balanceo. Mucho tiempo invertido pero seguimos adelante. Somos Corporación Robot