Larry Stevens

14.4K posts

Larry Stevens

Larry Stevens

@lfstevens

Not on your team

가입일 Şubat 2009
691 팔로잉248 팔로워
고정된 트윗
Larry Stevens
Larry Stevens@lfstevens·
Long past time for Tesla to flood the zone with data. They should let the entire fleet have FSD for free until they get the 10b miles or however much data to get the next 1-2 nines of reliability that they need. Time is of the essence and they could get there in no time if they chose.
English
0
0
0
14
Larry Stevens
Larry Stevens@lfstevens·
@VladSaigau Key question is whether they can build/operate a SOTA fab at scale quickly. If it takes 7 years....
English
0
0
0
0
Vlad Saigau
Vlad Saigau@VladSaigau·
Orbital compute is approaching cost parity with terrestrial data centers faster than the market thinks (~2028)... We built a bottom-up parametric model to find out when. TeraFab: Tesla, SpaceX, and xAI committed $25B to a 2nm fab in Austin. Elon already repositioned Dojo 3 explicitly for orbital AI compute. This is committed capital from the only entity that simultaneously controls launch, satellite manufacturing, chip design, and an AI customer simultaneously. Terrestrial costs are compounding... Datacenter construction inflation is running ~7% annually on the 60% of all-in CapEx that isn't GPUs. Current benchmarks: $35B/GW (Bernstein), $50B/GW (Stargate), $60B/GW (Jensen Huang). NIMBY, energy scarcity, permitting are driving costs up. Structural, not cyclical...these numbers only go up. We modelled three orbital deployment cost scenarios: Bear (possible ~2027): B200s at $45k/chip, $500/kg launch. Implies $132B/GW. Reaches parity with our most bearish terrestrial benchmark, Jensen, in year 7 of deployment. Base Case (~2028): AI5 at $5k/chip, $300/kg Starship. Implies $46B/GW. Below Stargate immediately on deployment. Below Bernstein by 2032. $24B/GW by 2038. Bull (~2030): D3 from Terafab at $2.5k/chip, $100/kg full reuse. Below every terrestrial benchmark from day one and ~12.5B/GW within a decade. Chip cost (relative to performance) is the most decisive variable. At $45k per B200, it would take 7 years to reach cost parity. At $5k per AI5, it inflects. At $2.5k per D3 from a captive fab, it breaks wide open. After chips, thermal hardware shows the widest sensitivity in our model, but only because there are zero public vendor quotes for space-grade deployable radiators at scale. It's a pricing uncertainty, not a physics problem. Wright's Law front-loads the advantage. Typical aerospace 15% learning rate on manufacturing and launch. Steepest declines in years 1-4 as Starship cadence can ramp, which is important for investment timing. The moat is extremely clear. SpaceX controls Starship (launch), Starlink (data transport), Terafab (silicon at cost), xAI (captive demand and software engineering). Fully closed loop. CapEx structurally insulated from commercial markups. Nobody else can build this. Orbital won't make terrestrial obsolete within the decade. But as ground-based bottlenecks compound, it becomes the marginal scaling pathway for AI compute infrastructure. The crossover is closer than you think. Full analysis 🧐: research.33fg.com/analysis/the-e…
Vlad Saigau tweet media
English
5
13
100
27.4K
nextbigfuture
nextbigfuture@nextbigfuture·
Tesla Semi saves money even before it become a robotruck. As a robotruck it can drive 3 times as many miles and without paying the drivers. @elonmusk
Josh Kale@JoshKale

The Tesla Semi might be the most important product Tesla has ever built. This is a huge deal. Any company buying a diesel semi truck today is lighting money on fire: - The Tesla Semi costs 15 cents per mile to power - A diesel Class 8 costs 48 cents That’s a 70% reduction in energy costs per mile… Over the battery’s rated million-mile lifespan, that gap compounds into $330,000 in energy savings, PER TRUCK Buy ten of them and you’ve saved $3.3 million. Buy a hundred and you’ve restructured your entire P&L. And that’s just fuel. The Semi has no engine, no transmission, no oil changes. Regenerative braking means brake pads last dramatically longer. Tesla’s pilot fleet is running at 95% uptime, and when trucks do break down, half are back on the road within an hour. PepsiCo, DHL, Frito-Lay, ArcBest, and Saia are already running them. 13.5 million real-world miles logged across the fleet. One single truck has done 440,000 miles. At $290,000, the Long Range version is also cheaper than the industry average of $435,000 for a zero-emission Class 8. At some point this stops being an environmental discussion and becomes fiduciary negligence for anyone still signing diesel purchase orders.

English
1
7
70
2.8K
God of Prompt
God of Prompt@godofprompt·
An independent researcher in Shenzhen just made your company's six-figure optimization software look like a toy. 710% gap vs 0.34%. No feasible solution vs solved. PhD required vs plain English description. > Every logistics team is overpaying for software that a GPU and 50 lines of Python just replaced. > The test was simple. Give every solver 60 seconds on a 150-node routing problem. The kind of problem every logistics operation runs daily. > SCIP the enterprise MIP solver your operations team probably uses finished with a 709.9% optimality gap. CBC, the other industry standard, couldn't find a single feasible solution. Not a bad solution. No solution at all. > cuGenOpt, built by one independent researcher, finished at 0.34% gap. On a consumer GPU. In the same 60 seconds. > That's not a marginal improvement. That's a different category of performance entirely. > The reason enterprise solvers fail isn't a secret. MIP formulations require O(n²) variables. At n=150, the mathematical model itself becomes the bottleneck before the solver even starts searching. The vendors know this. They've known it for decades. The answer they sold you was "hire better operations researchers." > cuGenOpt's answer was different. Run thousands of parallel searches simultaneously on GPU cores. Let the hardware do what the math couldn't. And make the whole thing installable in one line. → SCIP on 150-node TSP: 709.9% gap after 60 seconds → CBC on same problem: no feasible solution found → cuGenOpt on same problem: 0.34% gap, same hardware, same time limit → Custom constraints like intra-route priority ordering: 10 lines of code → Nonlinear cost functions enterprise solvers can't model: 5 lines of code → Describing your problem in plain English to get a working solver: now possible The part nobody is talking about: specialized enterprise solvers can't handle custom constraints either. Try adding "high-priority customers must be visited before low-priority ones within the same route" to OR-Tools. It can't model intra-route partial orders. cuGenOpt handles it with 10 additional lines. One researcher. One GPU. One pip install. The optimization industry just got a 30-year problem solved in public.
God of Prompt tweet media
English
7
30
251
17.5K
Vlad R P
Vlad R P@Sasha_rp88·
AUSTIN is 42 cars. On purpose. It's the only city with unsupervised rides. Every mile there is writing the safety case that becomes the legal template for every city that follows. One incident in Austin doesn't just hurt Tesla. It hands ammunition to rewrite AV rules for the whole industry. "Elon doesn't need to lobby for favorable AV legislation — he just needs to not give anyone a reason to write unfavorable legislation."
English
2
0
0
156
Vlad R P
Vlad R P@Sasha_rp88·
Nobody is connecting these dots on $TSLA robotaxi. The "slowdown" isn't a failure. It's the most deliberate strategy in Physical AI deployment right now. 🧵 #TSLA #tsla
Vlad R P tweet media
English
2
1
7
508
Larry Stevens
Larry Stevens@lfstevens·
@revishvilig Putin's model of recruiting/sacrificing from the provinces may have the long-term consequence of disconnecting those provinces from "Russia". Nice!
English
0
0
0
4
Giorgi Revishvili
Giorgi Revishvili@revishvilig·
Ukraine has pursued a strategy of bringing the war back to Russia, aiming to erode the perception that the full-scale invasion is distant and cost-free for Russian society. An analysis suggests that strategy is beginning to reshape how the war is felt across Russia. 1/13
English
9
148
1.3K
84.3K
Larry Stevens
Larry Stevens@lfstevens·
I sure hope that Elon is right about an imminent growth explosion. Otherwise, the world is in a big debt hurt that it has no way to address.
English
0
0
0
4
Larry Stevens
Larry Stevens@lfstevens·
@mattyglesias @grok what fraction of parents care about sending their kids to a school "skewed to a higher SES group of kids" to the point where they do something about it?
English
1
0
0
8
Matthew Yglesias
Matthew Yglesias@mattyglesias·
Schools can do a better or worse job of educating the students they have, but parents who care about school quality mostly just want to send their kids to a school with above-average peers which is clearly not scalable. slowboring.com/p/ed-tech-is-n…
Matthew Yglesias tweet media
English
39
19
219
43.6K
Larry Stevens
Larry Stevens@lfstevens·
@LimitingThe Faster than anyone thinks possible/has ever been done before, and yet way behind schedule. Worried that those delays will compromise everything else. Elon is not usually a man with a backup plan.
English
0
0
0
1
The Limiting Factor
The Limiting Factor@LimitingThe·
One of the most exciting parts of Elon entering the chip fab business for me is that we're going to discover the relative difficulty of a chip fab compared to launching rockets and manufacturing batteries 😁
The Limiting Factor@LimitingThe

TSMC mskes thousands of different chips for hundreds of customers. Terafab will make two chips. This massively simplifies the factory and allows them to focus on trying different iterations of those two chips to accelerate production and reduce costs.

English
53
67
1.5K
100.5K
Ramez Naam
Ramez Naam@ramez·
There's a real possibility that we've reached the turning point in the Ukraine war. Ukraine is: 1. Taking back territory. 2. Outgunning Russia in deep strikes. 3. Manufacturing it's own Tomahawk-class cruise missiles (with Europe's help). 4. Killing Russian soldiers faster than Putin can replace them. There is still a lot of war ahead, unfortunately. And factors like the surging price of oil help Putin. But the tide appears to be turning. Ukraine will grow stronger every month as its modern drone and missile capabilities continue to grow. Putin should sue for peace. The deal he'll be able to get will just get worse and worse from here.
Anton Gerashchenko@Gerashchenko_en

Global attention is focused on the Persian Gulf. But we have news too. March 2026 became the first month when, in terms of deep strikes, Ukraine reached parity with Russia. The average number of long-range kamikaze drones launched by Ukraine has significantly increased, while the average number of Russian strike drones launched against Ukraine has currently dropped to 90–110 per day. Experts offer various possible reasons for this decrease: ▪️ drone production in Russia was not fully localized and partially depended on Iran; ▪️ logistics routes have become longer, and Iran itself now needs drones; ▪️ Ukraine is partially hitting production sites and factories; ▪️ possibly, Russia is stockpiling drones, perhaps for hybrid provocations in other regions; ▪️ it is also likely that Russia has started supplying drones to Iran. By spring, Ukraine has significantly increased strikes on Russia using long-range kamikaze drones. For example: on the night of March 18, Russia officially claimed to have "shot down" 238 drones; however, some of them struck a chemical plant in Stavropol Krai and an electronics facility in Sevastopol. Ukraine has also expanded its operational range: over 1,500 km. Russians have openly expressed concern about the security of military infrastructure in the Urals. Ukraine has increased and continues to increase both the production and variety of drones. Moscow has been declared a target: from March 14 to 17, Russia reported about 250 drones in the Moscow air defense zone. 📹: Ukrainian Defenders destroying Russian air defense systems in Bryansk region of Russia

English
226
1.4K
7.4K
282K
Larry Stevens
Larry Stevens@lfstevens·
@grok has Tesla persisted and finally solved the dry coating task for its anodes and cathodes? Isn't that an example of Musk's persistence ability to surmount seemingly impossible obstacles? Isn't Terafab an example of Musk's ability to properly define a problem based on first principles?
English
1
0
0
10
Fred Lambert
Fred Lambert@FredLambert·
This is Tesla’s Battery Day on steroids. And if you’ve been following how that turned out, you should be very skeptical. In September 2020, Musk stood on a stage and promised a revolution in battery manufacturing with the 4680 cell. Tesla was going to ramp to 10 GWh within a year and eventually reach 3 TWh by 2030 — enough for 20 million cars annually. The dry electrode process was going to cut costs by 50%. Five and a half years later, the 4680 program has been a disappointment. Tesla’s own top battery supplier said Elon doesn’t know how to make battery cells. The dry electrode process needed six or seven revisions. It took years longer than promised, and the 3 TWh target is a distant fantasy. Tesla is estimated to be at only about 2% of its original cell manufacturing volume goal. Now Musk wants us to believe he’s going to build a chip fab. Not just any chip fab — the biggest in the world, at 2nm, producing 70% of TSMC’s total output from a single building. Battery cell manufacturing is difficult. Chip fabrication at the leading edge is on another planet of difficulty. TSMC spent $165 billion over years to build six fabs in Arizona, and those won’t reach 2nm production until 2029. A single 2nm fab with 50,000 wafer starts per month costs roughly $28 billion, and it takes about 38 months just to build in the U.S. Tesla has zero semiconductor manufacturing experience. The timing tells the real story. Tesla’s auto business is in freefall — sales declined for the second consecutive year in 2025, with a bloodbath in Europe and its first-ever annual decline in China. SpaceX, by contrast, is about to IPO at a potential $1.5-1.75 trillion valuation. This announcement is clearly designed to attach Tesla, a business in decline, and SpaceX, a business about to go public, to the AI hyperscaler narrative, a boat Musk has already missed with xAI, which he admitted “was not built right” and had to be bailed out by SpaceX. And the cherry on top, or in space, rather, is the plan to put 80% of this compute in orbit. Data centers in space. Powered by solar panels. Launched by Starship. This is the kind of vision that sounds impressive on stage but has essentially zero connection to any near-term business reality, or any possible reality at all, according to most credible experts. The whole thing reeks of desperation. Musk is hyping an 8th-gen AI chip while he still hasn’t delivered on the promises made with the 3rd generation. He’s promising to do in a couple of years what TSMC has spent decades and hundreds of billions of dollars building. We’ve seen this movie before with battery cells, and we know how it ends. electrek.co/2026/03/22/tes…
English
72
109
682
72.6K
Jim Park🏀🌌
Jim Park🏀🌌@Sheridanblog·
FINALLY. It's been torture watching without Curry in a lot of games. Can't be back soon enough. Next step is to see where Horford and Seth are at. Horford's defense puts them at a different tier and we never got to see Curry run with him as the starting C. Another thing that's yet to happen is Curry-Curry combo. Kerr the clown POS ran all the trash 4G combos but couldn't bother to allow the Curry bros to play a single minute together this season. Real scumbag work, Curry, Moody, Santos, Horford, KP. We need to see a starting unit like that and see what this team looks like. Seth, Melton, GPII, Draymond, Post. This is the vision for play-in/playoffs Cryer/Williams/Leons should be depth pieces with Podz being a very distant option, he will suck when it matters most. This team can still peak just at the right time to shock everyone. I will not take another Curry year for granted, talking about "save him for next season". Nothing is guaranteed.
Anthony Slater@anthonyVslater

The Warriors have ruled Steph Curry out of the next two games (tonight and at Mavericks on Monday), but he will be incorporated into live practice in coming days. Evaluated again when they return home on Tuesday. GSW starts homestand on Wednesday vs Nets.

English
9
1
59
8.2K
Larry Stevens
Larry Stevens@lfstevens·
@seti_park Each object needs a bunch of things to characterize it: - Object type (o) - Position (x,y,z) or (r,d) - Size (x,y,z) - Direction of movement (r) - Speed of movement (v) Complicated
English
0
0
2
453
Larry Stevens
Larry Stevens@lfstevens·
@LimitingThe @grok Competition will drive prices down, shrinking GP. Supply will expand massively. Consumer surplus will be the big winner.
English
0
0
0
4
The Limiting Factor
The Limiting Factor@LimitingThe·
What net present value does this indicate for a that company captures 50% of the US Robotaxi market by 2030 @grok?
Brett Winton@wintonARK

Let's dimension the US robotaxi market (since market participants seem unwilling to do so). People pattern match against structurally ~$3 per mile point to point mobility products and so misunderstand the potential scope of robotaxi as it becomes mass accessible. The average US adult spends nearly an hour per day driving. The imputed labor cost of all that manual piloting runs in excess of $4 trillion per year. In addition we pay $1.6 trillion annually for the actual service of driving point to point. By giving people back time (for which they don't have to pay full freight) and winning spend share, we think the US market could approach $4 trillion annually at saturation. Given reasonable expectations of supply diffusion and consumer adoption robotaxi service providers could exceed $1.5 trillion in revenue by 2030 with gross profits in excess of $1 trillion.

English
4
2
57
13.4K
Dr. Ben Tapper
Dr. Ben Tapper@DrBenTapper1·
AI data centers can use up to five million gallons of water each day. Am I the only one concerned about how this could pose a serious threat to our farmers and our food supply? What happens when these centers seriously strain our aquifers and dry up irrigation systems? Is it really worth it?
English
1.5K
2.4K
6K
105.3K
WW3finalboss
WW3finalboss@WW3finalboss·
🇺🇦🔥 ZAPORIZHZHIA ENCIRCLEMENT SHOCK Russian sources are now conceding that an entire brigade was surrounded in Zaporizhzhia as Ukrainian forces maneuvered decisively on the battlefield, with AFU advances (dark blue) tightening the noose and exposing critical vulnerabilities in Russian positioning. Momentum is shifting faster than expected—this could ripple far beyond the front line. @actfast
WW3finalboss tweet media
English
70
771
4.6K
181.6K
Larry Stevens
Larry Stevens@lfstevens·
@tegmark Our education system is so bad that it leaves unbelievable potential out of reach. If we fixed it (money is not the issue) our society would advance beyond anything we can imagine.
English
0
0
0
1
Larry Stevens
Larry Stevens@lfstevens·
Warriors are my team, but their offense is so bad. Fumbled passes, bailing on shots, inability to attack the rim (bright spot: layup drop-offs - GPII), 23% threes last night, 72% fts, team needs a reboot. Not seeing flashes of brilliance from any of the young guys. Glad they dumped JK, but he did show those flashes. Steph is all-time great, but he can't bench-press this team. Dray needs to go network. Time to tank.
English
0
0
0
3
Dr. Eric Berg
Dr. Eric Berg@dr_ericberg·
Here's something that might surprise you: The recommended daily intake for vitamin D, which has been cited for years as 600 IU, is based on a massive error. In 2014, researchers reanalyzed the original data used by the Institute of Medicine and found the actual necessary amount was closer to 8,895 IU! This recalculation has been confirmed by independent research, yet it is still completely ignored. This isn't just a "mistake," it's a systemic error that continues to leave millions deficient in a crucial vitamin, impacting immunity, mood, cognitive function, metabolic control, and even cancer risk.
English
147
826
3.4K
312.5K
Joel C. Sercel, PhD
Joel C. Sercel, PhD@JoelSercel·
The secret plan is out! Two great articles today about our plan to capture a small asteroid and bring it into Earth orbit. The concept is part of our New Moon mission, which explores relocating a small near-Earth asteroid into a controlled orbit as the first step toward building industrial infrastructure in space. Our team has been developing the key technologies for this approach for years, including capture mechanisms to constrain and move small asteroids and orbital debris. Accessing materials already in space could eventually enable a new generation of industries beyond Earth. If you're interested in the future of space resources, these articles are worth a read. hashtag#space hashtag#asteroidmining hashtag#spacetechnology hashtag#spacex hashtag#venture
Joel C. Sercel, PhD tweet media
English
63
115
955
93.2K

탐색

@VladSaigau @nextbigfuture @elonmusk @godofprompt @grok @Sasha_rp88 @revishvilig @mattyglesias