I hesitate to recast the discussion in a negative tone, but this doesn't sound right on technical terms.

Is World Labs' Atlas genuinely novel? Their demos do not seem to be better than the existing state of the art.

Fei Fei Li has been criticized even in the ImageNet days as a shower than a doer. Her roadshow past two years extolling "world model" in vague terms didn't show much insight, and her company's demo seems to merely rehash what has been available in the field for years. If you haven't seen the current SoTA in Gaussian splats, then World Labs' demos may seem cool, but they are mostly standard in the field now. If someone is more familiar with the inner workings of World Labs, I am happy to be corrected (but inner details wouldn't change the fact that demos are no better than SoTA).

I really had high hopes for World Labs, but I am quite saddened to realize that perhaps all this was just a financial maneuver and the detractors were right from the beginning.

source: I used to be in the downstream field: Gaussian spats for robotics. The exact field that World Labs' is supposed to help.

I get the feeling that people still think World Labs is all in on splats? I agree the original splat demos are not competitive relative to the SOTA for splats. But Atlas (https://www.worldlabs.ai/blog/atlas) is more of a generative model than a splat-oriented model. They are essentially trying to combine many useful capabilities into one omni-modal generative model. For instance, native camera pose conditioning is not a capability most video models have, and it also seems to have some SLAM/VGGT like capabilities (see blog).

Essentially, this is the first true "omni-model" that can be conditioned on anything (text, pose, depth, video) and give you any output (pose, depth, video, splats). You could argue that they are not SOTA right now relative to a traditional multi-stage pipeline, but this unified model is much more amenable to scale.

edit: they also seem to be SOTA for 3D reconstruction (better than VGGT-Omega, DAv3, pi3, which I think is very solid).

which if you know why fei fei is famous, perfectly lines up. she was the first to turn from better algs to diverse data for training, even for super specific image classifiers. what is this, but an attempt to, like humans, make anything we can experience training data and thus a native "language" of the model.

I work in a highly related area. But honestly:

> has been criticized even in the ImageNet days as a shower than a doer

When has the world ever rewarded doers? It has almost always rewarded the showers. If it rewarded the doers:

* CTOs would be regarded more highly than CEOs

* A few key Apple engineers would be more well known than Steve Jobs

* Nikola Tesla would be more famous than Edison

> this was just a financial maneuver

I mean, if you could have a couple billion dollars now OR possibly $0 when OpenAI/Anthropic's models can eventually do what yours can do, I wouldn't blame you for taking the former.

ignoring op's out-of-place personal attacks, what about the world lab's model capability - do you agree?

I think the model is legit but not worth 8 billion dollars, nor is there much secret sauce. I'm at a series C company and I think we're pretty confident we have the resources to reproduce it internally.

You say the world doesn't reward the doers but CTOs and a few key Apple engineers were/are rewarded with heavy bags of money. They don't get the media features because the company has to pay the media to feature Steve Jobs telling us he's just a cool hippie, actually, and not a ruthless psychopath. Doing the same for the CTO serves no marketing purpose for Apple.

As for Nikola Tesla, he spent a lot of his money on things that didn't work out, like the unfinished Wardenclyffe Tower.

None of the examples you shared is the world not "rewarding" them compared to people who are not those things.

Doers need to spend a lot of resources on things that don't work out. It's normal that only 1 in 10 things work out.

The showers usually pluck them out and show them, and look like constant success to the public.

If the World Labs people are talented, it could be just an acquihire right?

8B/200=40M (avg) per acquihire. you really think those people were worth that much?

AMD has made a lot of acquihires in the last few years. this ain't one of em.

Correct, this is an acquisition and will benefit AMD a lot. In fact, AMD has made some great acquisitions lately and are on a roll.

I don't think they published details but the Atlas model is a feed forward neural network, and not Gaussian splat based model. Unlike GS it works from a few images.

Gaussian splatting is the rendering technique of projecting the gaussians into the 2D viewport. How you obtain the gaussians in the first place is a different question (feed forward prediction vs. iterative gradient descent fitting).

Didn’t she freely release much of the industry-shattering work she did on which other orgs capitalized?

i'm quite happy with all the scientists and ml pioneers getting paid - but it seems like we've reached a point where a billion dollar is not a big amount anymore and that is just strange for me to accept. It used to be that being a unicorn is a big deal, now they buy companies without any products for billions ... Just leads me to believe that those at the top have just become unfathomable rich - be it individuals, vc companies or corporations.

To pay for WWII, the US had a 90% federal tax bracket that effectively capped yearly income. In the 1950s, a CEO could only earn the equivalent of 5mil a year in 2026 dollars. That is why vacation destinations used to boom. Why people could take road trips because work wasn't understaffed. How we funded interstates and modern infrastructure. It's why the one parent household existed. CEO's don't underpay, understaff, or outsource when it doesn't personally make themselves richer.

This is what mcarthyism was really about. A handful of rich people calling any taxes that apply to them a form of communism. That group has basically won. They repealed the taxes and regulations that ended the great depression and now we are entering another depression. This new one will last until we vote to bring back the federal income tax policies and post-1929 financial regulations of the 1950s.

This was an interesting overview. Thank you.

A minor correction: both the original Cheetah and Mini Cheetah motor designs were QDDs and were using BLDCs. The innovation was from using a off-the-shelf BLDC model as much as possible instead of a custom-design/modify one. Moreover the placement of electronics and gearbox was much more compact, leading to the modern day design. If any thing, the original Cheetah motors were actually simpler than those in Mini Cheetah.

Is your Qwen3.6 locally run on your laptop? What kind of tokens/s are you getting from your laptop GPU?

Qwen is running on my Mac Studio, an M1 Ultra 64gb. My harness (oh-my-pi) on my laptop is configured to use the models hosted on my local network, since it's just a MacBook Air 16gb and probably incapable of running anything useful itself.

I get about 45-55 tokens per second using Qwen with this setup. I could probably squeeze out more if I messed around with the settings, but I'm mostly using oMLX's defaults for the model.

I have Qwen3.6 35B-A3B on my laptop and it does 60 tokens/s

Could you share the specs of your laptop?

M5 Pro/64 GB RAM

The whole mirror analogy for angular facets breaks down when you realize airplane is a moving object and at some point in its trajectory, one of the facets is likely to be facing the radar (maximum cross section).

If I remember from my numerical analysis days, it was the ease of numerical computing that necessitated the facets. Think 3D graphics from the 90s. All polygons.

3d graphics is still mostly polygons, just much smaller?

Anyway, I'd imagine the part that is an integral over the surface becomes a finite sum, the post doesn't say but it seems like a reasonable guess.

My understanding is that this more of a tactical consideration than a technical one. When these planes were used in the Gulf war, they were absolutely detected by Iraqi radars; they were just detected too late to do anything about them. While these planes were approaching the radar stations, they were oriented in one relative direction so the radar waves could be scattered away from the source consistently. Source is the Ben Rich Skunkworks book.

Your answer shows why flat surfaces were acceptable for a bomber (with an asterisk) given the constraints.

Then, the question becomes why is the B-2 bomber (and latest stealth fighter jets) not angular like F 117.

Scientific computing has come a long way since the days of F-117. We do not need flat surfaces to scatter the beams away. Finding optimal curved surfaces that scatter the beams away is just a harder math problem.