@TSLA_inside_ I would like to say Spain, because their enormous testing, but with a corrupt socialist at the helm that might hate Elon, who knows.
Just look at Sweden and their hate for Elon and Tesla.
@Cyber_Trailer I agree with you on principle, but a problem with doing Model Y RWD Premium as default in Europe is that it is the only model that uses the 4680 batteries which is not even longer range than the AWD model and has various reports of bad charging curves.
I actually do not know the limits. I am just running on my Premium+ (SuperGrok) and have never hit any limits. There are 2 models available. Grok Build (500k context) and Cursor 2.5 (200k context). I never hit any limits on the Grok Build one, and do a compact about once per session on Cursor.
@Aarchimandrita@JOBhakdi there has to be 100's of real technical reasons they could point at for delaying a bit. many of them just "opinions" like "I'd like this stat to be that value before.."
Jag har den senaste veckan testat Grok Build och det är fascinerande för "en gammal kodare". Jag har gjort ett fullständigt personligt projekt (Firefox Addon) och resultatet var över förväntan.
Har dock lite WSL relaterade problem med terminalen som jag nu testar lösningar på med hjälp av Grok, men det slog mig precis hur jag behandlar Grok som en "person/kollega". Vi diskuterar problemen ihop och bägge kommer med förslag.
#GrokBuild
Tesla investerare här: xAI skulle vara del av Tesla, men de "unika" ingenjörerna (11st) ville ha startup med bättre uppsida så xAI skapades. Målet var att konkurrera med OpenAI, Google och Anthropic. De lyckades inte med det för de satsade inte på det enda som genererade inkomst (AI Kodning).
Ingen klår Elon Musk på att snabbt skala saker, så xAI blev den som snabbast kunde bygga enorma GPU-kluster.
Samtidigt körde han stenhårt med Teslas AI5 chip (möte varje helg i 6 månader) och de har nu lyckats med "tapeout" (skicka till produktion) av den, och har planer på AI6, 7, 8 mm samt även "Dojo 3" som kommer vara konkurrent med Nvidia.
Alla dessa pusselbitar gjorde att han formulerade en ny plan. Då han har röstmajoriteten i både SpaceX samt xAI så var en snabb integrering av dessa enkel.
mindre än 2 veckor senare så blev det enorm reorganisation på xAI. Ingen av de 11 "unika" anställda jobbar längre på xAI, och de har satsat på kodning (via Cursor).
Målet har alltså svängt snabbt och verkar vara att göra samma sak man gjorde med StarLink när kapaciteten för att skjuta upp saker med StarShip kommer explodera., Man måste ha något att skjuta upp (skalfördelar) och i detta fall blir det AI satelliter med chip från Tesla (AI6+ och Dojo3).
Under tiden då de inte längre behöver lika mycket GPU kraft så passar de på att hyra ut den (min spekulation: för att boosta intäkterna inför IPO om 1 vecka).
Slutsats: Det finns en plan, den ser bra ut. Men företaget är enormt övervärderat (min åsikt) för som Tesla investerare så VET jag att om Elon säger att det är klart om 1 år, så lägg på minst 5 år till.
@aaditsh I would like to remind people that Google owns (aprox.) 6.11% of SpaceX. That will be about a 10x from their initial investment at $1.75T valuation to roughly $100B - $122B. So it is essentially "free" for them.
OP is right, it wasn't his plan for the Colossus Superclusters. Remember SpaceX bought xAI in February, within 2 weeks he introduced the big "re-org" and now none of the starting employees (the reason xAI isn't part of Tesla) is working there any more.
His plan was to be better than Anthropic and OpenAI.
@Pirat_Nation I have not seen mentioned the memory bandwidth of the new chip. On the BD10 (DGX Spark) it was 279GB/s. This is the most important stat for local AI as it determines the Token/s speed.
For reference:
* M5 Max (600GB/s)
* RTX 5090 (1792GB/s)
I was suggesting that SpaceX "future value" was unproven because the things that has to work for it to be a reality is not there yet.
I am well aware that tesla will (in the future) need a lot more AI chips to scale Optimus and Robotaxi. Even though if we take Elons (Osborne, cough, cough) words for it that robotaxi won't need AI5 they have enormous orders for both TSMC and Samsung to make AI5 so they should be ok while Teslas Terrafab is built.
I will stop arguing about this now. I really like you Cern, but you miss that long standing Tesla investors like me want a "Show me the money" moment after all the promises from Elon, not more promises for a possibly bright future together with SpaceX. So in the end your well written, well meaning article does nothing for me. Sorry.
Starship is unproven (Yes, I stayed up late to see it).
Even if it can fly "ok" within a year, the heatshild has to work (unproven) so it can be reusable fast.
Terrafab is extremely unproven. Example: 4680 took at least 5 years extra time and is a long way from the promised 3TWh at 2030, and it is so much easier than Terrafab.
The AI sattelites are unproven, although I think this might be the easiest to get going.
The AI(x) chips are unproven, the AI 6-7-8 chips they plan to send into space is not even designed yet.
All these technologies are needed for SpaceX future value to come true.
Observe: I really hope they will, I am not against SpaceX and not a future merger. I just think it is madness to merge at this point in time, from a Tesla investors point of view.
You have to stop using this as you answer to every reply.
Yes, I read it, now point out what scenario values Tesla "a lot" more than SpaceX? (I don't care about SpaceX unproven future value).
Most Tesla investors can see Teslas future value because it is being manufactured right as we speak, but SpaceX future value is "unknown" and unproven. We all know how late things can be that Elon promises, so why should SpaceX be any different?
SpaceX has about 25% the revenue as Tesla yet is valued above Tesla. To suggest they merge as "equals" just when Robotaxis are rolling out and optimus V3 is presented in 1-2 months is madness in my mind, and specualtion about SpaceX future value is not going to change that. You need better arguments.
@bevedoni@karpathy He got "f**k you" money during his years in Tesla where he had to do what may have been boring but necessary.
Now he can choose where to work without having to worry about the pay, he can select a company that will let him do whatever he fancy, because his track record is great.