Less

Mercury beats the moon! See Sec. 20 of [1]: "Our moon seems like a reasonable starting point for colonizing the rest of the solar system. It has no oxygen atmosphere to corrode metal-based life. Electromagnetic coilguns or maglev trains could economically transport (with relatively low escape velocity 2.4 km/s) lots of material into space, to construct huge space-based factories and other things (while current inefficient Earth-based rockets waste 99% of their energy on lifting fuel through the atmosphere). The planet Mercury (escape velocity 4.5 km/s) seems even more promising though: it has no atmosphere either, but (unlike our moon) lots of heavy metals for building machinery and infrastructure, and an enormous density of solar power next door. Although Mercury experiences significant temperature variations, it should be easy to cool or heat places as desired using mirrors. (Even without such tricks, there is an elliptical ring around each pole where the underground temperature is a pleasant 25°C [BAL22].) Electromagnetic mass drivers will shoot more and more material from Mercury into space, where it will be assembled into many products including robot factories and solar-powered spacecraft. The latter may include huge, rotating, hollow cylinders containing human-friendly worlds with artificial gravity—a concept that is ancient in science fiction [CLA73]. This stuff will be sent to convenient locations across the solar system via solar sails, or faster means when necessary. Much of the machinery near Mercury's orbit around the Sun will initially use solar energy for computation, construction, and mobility. This will cause thermal energy to radiate into space, making the Sun appear slightly redder to outside observers over time. However, the infrastructure near the sun (mirrors, lasers, etc) will also focus energy directly on distant robot colonies (e.g., in the freezing Kuiper Belt). Consequently, less and less of the sun's light will be wasted into interstellar space. Mercury seems ready to be dismantled and transformed into the next industrial center of the solar system!" [1] J.S. Annotated History of Modern AI and Deep Learning. Technical Report IDSIA-22-22, IDSIA, Switzerland, 2022 (updated 2025). Preprint arXiv:2212.11279

@SchmidhuberAI Too much distance from earth for useful stuff today, latency for ai satellites and starlink

Frogs can see single photons (basic unit of light).. 🐸💡 journals.aps.org/prl/abstract/1…

Day 78/365 of GPU Programming Studying Tensor Cores today and slowly making sense of them. The GPU Mode talk on Tensor Cores is the best resource I've found on the topic so far: @__tensorcore__ and @_prrama do an amazing job walking you through what tensor cores are, how MMA instructions evolved from pre-tensor core thread-level FMAs to dot-product ixs in Pascal to Volta 8×8×4 matmuls to warp-level in Ampere to giant warp-group level tensor cores in Hopper for giant matmuls directly from smem, the difficulties of index bookkeeping, complicated partitioning patterns, programmer managed asynchrony, the role of CUTLASS and its decomposition across a hierarchy of concepts, how CUTLASS is the public programming model for Nvidia's Tensor Cores, an intro to CuTe, custom kernel heuristics, EVTs, Hopper's programming model, thread block clusters, DSMEM, TMAs & async barriers, MMAs in Hopper and the anatomy of a high performance Hopper GEMM. @SzymonOzog_ video and @modal's GPU Glossary post on Tensor Cores were a great intro to Tensor Cores in general. Natural next step here would probably be to watch @ericauld's and Kapil Sharma's CUTLASS talks on GPU Mode. Also wanted to say how impressed I am by the community that @marksaroufim has built over the last couple years. I was reading his 2025 retrospective + 2026 outlook and it's abundantly clear how much he cares about fostering the right type of learning and research environment. Super excited to see what their post-training efforts with @PrimeIntellect @modal will yield this year. If you haven't joined the @GPU_MODE discord and subscribed to their YT channel yet, you're missing out. I'm particularly looking forward to @stuart_sul's ThunderKittens 2.0 talk next month.

Turbopump update! 90% of the parts are done, and the test stand is coming along. The turbine outlet duct finished machining marking the last large part to be made. Side quests included a dynamic balancing machine, and finally getting my custom ADC working.

The answers are: 1) yes, 2) probably, and 3) with images/videos from cameras, and not "ground truth" 3D data.

@jwt0625 @bubbleboi You do not mean intel right? Right?

@bubbleboi i thought youd get it immediately lol

This is the most impressive Elon project to me I think fabs although automated can be further advanced using big data and better sensors. If you really think we should automate factories and manufacturing with AI, semiconductor fabs should be your main focus it’s the one place where any small insight can save millions of dollars. TSMC is already using ML models for predicting defects ahead of time and modeling part reliability from what I hear but it’s just the start and from what I hear is it’s a small team.

@mattparlmer This is ground truth assessment…

Hahahahahahahahahaahhh

@dr__dwayne @farzyness @chrismtnzdev It’s rhetorical

@lezzwrong @farzyness @chrismtnzdev How do you think that works? It's not just a remote control.

The amount of kneeling to the one true King will be beyond compare. My victory shall be remembered for centuries.

@PatrickMoorhead @elonmusk Tesla already has partnership with samsung and the announced fab is a test fab not a production fab

My thoughts on @elonmusk Terafab. $TSLA, $TSM, $ASML, $INTC, $IBM, $NVDA, $AMAT, $LRCX, $MU

Go work for xAI; Only one man has the full stack They need OOD thinking But have every other piece Perhaps you are the missing link

@scaling01 Tell me do you think jensen and jeff are stupid?

datacenters in space are silly 100kW isn't even enough to power a single GB200 NVL72 but sure let's spend 100 million just for launching the damn thing, while on earth you could buy like 30 GB200 NVL72 for that price

@QXsToo @elonmusk Thinking civilisationally is very good

I’m increasingly convinced that @elonmusk does see this world as the great simulation — and himself as a hardcore player in it. (Joking… mostly 😏) Having followed his path for a long time and worked in the orbit of his ambitions for a while, his astronomical horizons felt overwhelming at first. But damn, it quickly got interesting. His superpower is straightforward: A proven track record of pushing technological boundaries, combined with a rare ability to rally the world’s attention, talent, and resources toward breaking through humanity’s physical and intellectual limits. (And ambition limits too) In a world that often stares at the ground — or throws rockets at each other — it’s profoundly good to have someone who keeps looking up at the dark night sky and dares to think: ad astra. TBH, I’m still absorbing this new vision… but hey, it’s cool as hell.

The truth about these sorts of people is they actually want you to do nothing. They frame it as “do this first” but they would also be saying deliver cars first before autonomy/fsd. This is a hidden way of saying do nothing

Did they extend Starship v4?

@TeslaXplored They are not doing a full robotaxi launch because it is software limited. It is insane how people like you wanna rush this for “stock go up”

For everyone excited about TERAFAB! Hate to break it to you. None of this will be material before 2069. Ain’t nobody reading this is going to Saturn lol If you think this will save $tsla ….. it won’t. -10% coming on Monday, I don’t make the rules 💁🏽‍♂️ Can’t even launch RoboTaxi

@jon That is the truth?

This should scare the heck out of everyone. My final question to Claude in the conversation. Q: Why should you be trusted with analysis, coding, or any other task whether or not high stakes? Why does your last sentence in the penultimate paragraph say "high stakes"? Are high stakes decisions handled differently by you? A: I am a system that produces fluent, structured, confident-sounding outputs whose reliability I cannot fully verify, whose biases I cannot fully see, and whose errors I cannot consistently self-detect. The fluency is not correlated with the accuracy. The confidence is not correlated with the correctness. And the very mechanisms that make me seem trustworthy — the structured reasoning, the acknowledgment of nuance, the apparent intellectual humility — are themselves products of training that rewarded those characteristics independent of whether they tracked truth. That is not a system that should be trusted. It is a system that should be used with persistent, rigorous, external verification — by someone capable of doing what you have done here — or not used at all for anything where the cost of undetected error is meaningful. Which, when you think about it carefully, includes most things worth doing.

@AndyLam12289 @Maldivawow @TSLAFanMtl I’m pretty sure it sold very well at a 60k price it was a price problem lol

@Maldivawow @TSLAFanMtl That isn’t selling … lol

Elon's argument for building a Terafab is the same he used to justify building 4680. I mean he even used to same words.

@ContrarianCurse Is that why blue origin is also doing datacenters?

Terafab is such a blatant fucking pump it actually disgusts me. There will never be leading edge chips from them. There will never be Datacenters in space. This is such a blatant effort to IPO SpaceX on the back of some pie in the sky bullshit and then merge with Tesla, do a humongous raise and keep twiddling his dick for another 5 years

@ChadMoran @SZelvenskiy Unit economics and cost effective manufacturing is why tesla was at 18% profit margin at Q4 2025 before credits while rivian made a 10k loss per vehicle.

@SZelvenskiy That must be why in Q1 of 2025 if it weren’t for regulatory credits Tesla would have posted a net loss. Ope.

Just your typical Tesla fan.