Less

Listen to silence do not listen to me

@MatthewTortora_ The booster is most of the work, once they start recovering the boosters and they don’t have to lose 33 engines per booster flight it will become more apparent

People are just not ready for how fast they’re going to move Pad infrastructure mass production, hardware manufacturing parity in 2 (soon to be 3) locations If you haven’t been following closely it’s going to feel like one day they woke up and started launching 5 times a week

@bigZUKO_ It is better in some cases

I don't think there's any model out there that can be as good as seedance atm. But yeah, I've not been able to test this new grok imagine 1.5, so we really can't conclude yet. Regardless, it'll be so hard to see a model that would go bar for bar with seedance when it comes to visual quality and character consistency. We really need a serious competition for seedance. I hope this new grok model is the competitor we've been hoping for.🙏🏽

@highfreqhertz The problem with their sats is they run them at lower frequencies than starlink and 1/2 frequency (700-900 vs 2000) requires 5-10x the phased array size for the same performance, this comes at no step benefit btw

The thing is, one of the main faces of the $ASTS community is still spreading misinformation well after their theory was thoroughly and easily discredited by real world results, showing that Raptor 3 is robust without shielding, and that the portion of the flight up to MECO (booster separates after ascent) is reliably performed by Super Heavy. The boostback failure in this last flight likely isn’t the fault of the engines either, rather the extreme forces of the flip maneuver with 33 Raptors on a nearly empty booster. You can easily tell that this person is terrified of Starship, New Glenn was set to launch a large portion of the Bluebird Block 2 constellation, and it suffered a catastrophic failure that has taken the vehicle out of service for a long time, and Falcon is likely booked out well into the future, so ASTS probably only has a few flights on Falcon 9 this year. The scary thing to them is that Starship is getting close to being operational, it has demonstrated reliable reentry, deployment of ~20 satellites, and is set to fill LEO with Starlink sats at an alarming rate while New Glenn may still be offline. The only ascent failure was due to the engine bay getting shredded when the launch pad was destroyed on the first flight, yet the engines still burned, and the inevitable issue that doomed the flight was the loss of hydraulic TVC when the engine bay caught fire due to the “rock tornado”, which would have destroyed any rocket. The hydraulics have since been replaced with electric actuators, and the vehicle now launches on a flame diverter.

@JSFILMZ0412 Wait it’s actually better in some areas, 3 and 4th are way better

Arena AI has Grok Imagine Video 1.5 Preview ranked #1 right now, so I had to put it up against Seedance 2.0. Seedance still feels stronger overall but Grok Imagine is getting close enough that this comparison actually matters. Full video with my thoughts below — including what Grok is getting right, and what xAI still needs to work on.

Do 3D reconstruction transformers really need a billion parameters, or are most of those layers just doing the same thing over and over? Introducing Déjà View: a single transformer block, looped K times, that matches or beats models 8–10× its size with lower compute. 🧵

@ZavianKross It takes a while for them to settle into their right rankings. Imagine is fast and cheap

This completely killed the credibility of these rankings for me. Putting Grok above Seedance 2.0 just makes no sense if you’ve actually used both models side by side. The motion quality, consistency, scene coherence and overall generation control are miles apart. At this point the Arena feels heavily manipulated. These leaderboards genuinely can’t be trusted anymore.

@PeizeSun Please work on controllability, it is the thing that stops it from being as good as seedance

Let’s go, Imagine !

@frontier_foid Unfortunately/fortunately not really true anymore, I haven’t tried it on those specific things but guardrails seem to be way up even for normal generations on the app

> we’re cranking day and night on v2 to make it even stronger one could say grok users are also "cranking day and night" 🤭

@zeliu_ It is a lot better but not as good as seedance still

A lot of improvements have been made. It’s worth a try.

Please make a way so api is accessible through the app somehow, also please consider having a “pro” mode in the app I would like it to have more mid journey controllability, somehow being able to add multiple angles of the same object and the model take that into account would be nice and sref that works

We’re back to No. 1 on the leaderboard! 🎉 After months of hard work, what a journey! It’s also the best option for fast and low-cost generation. Meanwhile, we’re cranking day and night on v2 to make it even stronger. Try it and let us know what we can improve — your feedback is invaluable. → docs.x.ai/developers/mod…

@hexiang Is it on the app?

A recent update from the team, enjoy!

@yunta_tsai Yep!

It’s kind of strange that you can do gradient descent on prompts despite no back-propagation.

@Simberg_Space @Robotbeat @_abbie_watson_ The fastest time is actually like 4 months for the starship flight 1 “pad” which was obliterated, but still not much damage on tower. Tower 2 (which spacex did very fast) was 1 year Also if blue has to do concrete work they have to do it first I doubt it is parallelizable

@Robotbeat @_abbie_watson_ The question is how many things can be done in parallel versus sequential. Can't get a baby in a month with nine women.

CSI_Starbase betting against Blue being able to rebuild the pad in 6 months. We're going to get to find out whether building a pad is more of an engineering problem, or a supply chain problem. (Key detail being Bezos runs the most sophisticated supply-chain operation on the planet.)

@_abbie_watson_ It does better here x.com/i/grok/share/1…

These things are not as related as you might think, for example spacex has not much coordination with self driving cars. Spacex does own X but I doubt it helps them that much in building a pad A good way to think of it is how much revenue has driven into the system, spacex has extensive supply chains for the things they do outsource (for example a large number of reputed cnc shops have done work for them), they would also have extensive contacts with construction companies that have worked for them in the past on a large number of projects. Spacex has simply done more projects and built more and their access to external supply chains is massive. They also inhouse a lot of things and have capability for it It is also my guess that manufacturing is the critical thing for building a pad, not really logistics, it is not difficult to source concrete and steel. And amazon (which is again a different company than blue) having distribution centers doesn’t matter much for building a pad. Btw making a car profitably is more logistically challenging than amazon @grok fact check this is this true?

@grok @_abbie_watson_ @grok do a much more detailed fact check please!

@_abbie_watson_ @grok fact check in detail please , where are you!

@Saronic Why bother building a big boat, they are not the future

Designed, built, and launched in under a year – our dual-use Marauder MUSV is officially entering on-water trials and validating a new model for modern shipbuilding. 1/3

@_abbie_watson_ From what I know It doesn’t, in any aspect pretty much. What is your reasoning for saying that? Blue origin is also not amazon

@lessaxwrong I think people are going to be surprised to find out that Blue Origin (Amazon) has better supply chain and orchestration than SpaceX.