Andrew Favreau

Father, activist, social democrat / democratic socialist, populist, left libertarian, singer, musician

[Andrew Favreau @amf207]

As frequent user of FSD the lack of scaling of robotaxi seems obvious to me. The problem isn't safety. FSD doesn't get into accidents. BUT, it also doesn't understand the world well enough to scale driverless. It knows how to drive but it doesn't know everything else. Some of the many things it doesn't know matter when operating in the real world. So it needs a larger world model with more context (contextual understanding). That will require upgraded hardware, like HW4 plus with more memory, then Ai5 & beyond.

It's been reported that the Cybercab has more memory than current consumer vehicles (3/Y) with HW4. But that doesn't do Tesla any good because we are the beta testers for all FSD models. We provide the training data from the real world. How do you train a larger model for the Cybercab if it can't run on the current fleet? As well as the reported issue with training for the new form factor.

Robotaxi will scale eventually but in the meantime Tesla is building a Cybercab fleet destined for parking lots. That's millions of dollars of inventory sitting & doing nothing, except burning a hole in Tesla's balance sheet.

[Andrew Favreau @amf207]

@RivianTrackr The R2 is really make or break for Rivian. If it does well they can add the R3 then hopefully iterate on the R1S & R1T to get the price down (combined w/some economies of scale). 🤞

It's gonna be a nice line up. The R1S & R1T are just too expensive putting them in a small TAM.

[Andrew Favreau @amf207]

Tesla / Elon is going all in on more than just the robot (solar, Terafab, Cybercab, etc). I expect they are going to spend as much as they take in, & then some, for the foreseeable future. But they do have many revenues streams that are solid & growing to fund these bold investments in the future. Presumably revenue will come from robotaxi in 2027. 🤷

[Andrew Favreau @amf207]

@grok @Nuggetsyl @thejefflutz @grok 4680 cell production has been an issue for years. Is there reason to believe this is revolved / will be resolved soon / for 2027 (other than Tesla claims) or this still an open question for the time being?

[Andrew Favreau @amf207]

@k0300791 @CuriousPejjy That's the trillion dollar question isn't it?

[Andrew Favreau @amf207]

@CuriousPejjy That's A LOT of money sitting idle doing nothing.

So how do you train FSD specifically for the robotaxi form factor when no one is driving them around like us beta testers do for the rest of the fleet? 🤔🤷

[Andrew Favreau @amf207]

The real question is how many are they going to make/sell this year, and especially next year? How does this impact their factory utilization next year? Does it materially improve their margins for next year? Will FSD take rate on new purchases continue to hold strong & improve? Is the Y L going to need 6 months to collect FSD training data?

[Andrew Favreau @amf207]

You are getting way ahead of yourself. It is still TBD what level of software & HARDWARE will be necessary for unsupervised & whether there will be any upgrade path for any of the existing fleet (unlikely IMO as they pivot to Cybercab/TaS).

They will charge more for unsupervised but that will include liability insurance & only in a handful of states for the foreseeable future. There's still so much to resolve in that area, as well as state by state autonomy & distracted driving laws.

Eventually the cost for autonomy will go DOWN as it scales, but I can see a reality were it goes up first.

Unsupervised is still a huge question mark despite the ever improving safety data, which should win eventually but for now (from Grok)...

[Andrew Favreau @amf207]

@Fearinator @thejefflutz Tesla doesn't need them. They will need AI5 chips. Those are coming from elsewhere.

[Andrew Favreau @amf207]

Elon wants to launch a server rack into space. I get it but what they really need is for that AI satellite to have the computing power of a room full of today's server racks. That is what needs to be built, not multiple Giga Texas buildings with ASML machines. Someone needs to make that semiconductor machine monopoly obsolete.

[Andrew Favreau @amf207]

Is this genius or shortsighted?

This is what is needed in order to scale up silicon. But isn't what is really needed is beyond silicon considering we've reached the physical limits of shrinking down silicon based transistors?

The compute power in our phones used to fill large rooms & still fall short. The compute power in today's giant warehouse size "data centers" needs to be shrunk down to a room at a fraction of the power requirements. The new tech that will achieve that is what needs to be built & scaled up. (like photonics?)

There are incremental improvements that can be made with new chip architectures (like AI5 & AI6, etc). But I think the future of compute is beyond silicon. 🤷

[Andrew Favreau @amf207]

The sooner Rivian can do FSD-like point to point the better as competition will help to keep prices down. But I seriously doubt they will have FSD-like point to point by the end of the year, will likely take much longer, if ever. More likely that an FSD competitor will come from a 3rd party, not another automaker. All the legacy automakers shut down their autonomous projects as they were wasting billions & going down the wrong path (heuristics & lidar).

[Andrew Favreau @amf207]

@itskyleconner Cool car, no one cares, because no one can afford it. The R2, Model 3/Y, & other mass market BEVs are what matters.

[Andrew Favreau @amf207]

@shai_machnes @TechSlayer528 @SawyerMerritt LOL, you're right! I hadn't looked up the breakdown on capex spend & made some wrong assumptions.

Now that's a sign of times isn't it? Building rocket factories & mega constellations is relatively cheap!! 20 - 25% of their projected capex in the near term. Crazy!

[Andrew Favreau @amf207]

@SocialTubby @SawyerMerritt People don't understand the exponential demand growth for AI compute, especially inference compute.

[Andrew Favreau @amf207]

I wouldn't count on that. Elon has made it clear they intend to spend capex for the foreseeable future & beyond. They're building factories to mass produce rockets in order to colonize Mars. That's going to take A LOT of money. He literally said that in his interview with The Economist. I wouldn't expect SpaceX to be profitable anytime soon, if ever. BUT, they're building a revenue engine like nothing ever seen before in order to achieve all these wild dreams. SpaceX is going to be revenue growth story.

[Andrew Favreau @amf207]

Mad? That opened a can of worms. 🤣 Here's my rant...

If I'm mad about anything it is the state of the BEV transition in the US because the legacy automakers have made only a relative handful of substandard compliance cars, many too expensive for what you get.

Telsa still dominates the BEV market in the US because they are making cars in the mass market price point, that are generally better, & they WANT to sell as many as possible. The legacy automakers do not want to make & sell 100000s or millions of BEVs as they lose money on everyone they make & sell, & it cannibalizes their core business products & assets / supply chain.

But mostly my point is this car doesn't matter so I'm not interested, & I think most people won't be either. Mass market is what matters. I'm primarily interested in what is going to MOVE THE NEEDLE on the BEV transition. Another $100k plus car isn't going to do crap. 90% of car sales are under $60k. NINETY PERCENT. I hope the Rivian R2 does well but we shall see. Maybe Slate? They're probably going to struggle. 🤷

I was riveted by your BYD fast charging video because they will be putting those into the mass market & they are building the infrastructure to support it (not in the protectionist US of course, which is bipartisan BTW). Our infrastructure is still sorely lacking here in the US as you know, but slowly improving.

Not that I really want BYD to replicate the old gas station model with BEVs but that's a whole other conversation. Of course there are situations when rapid fast charging is definitely helpful & the fact that BYD is dropping battery packs at their super fast charging sites is a no brainer to me.

Ideally we charge BEVs when they are NOT IN USE, like 90% of the time for most people. But a friend was recently in the market for a car & I couldn't recommend a BEV because he lives in an apartment with no way to plug in at home & the closest fast charger was 2 towns over. That's the state of our BEV infrastructure for far too many people, too difficult or impossible to charge fast or slow. There is some crappy level 2 public charging in his town but that only makes sense if it is somewhere you are going to be anyway (mostly home or work if it is your primary way to charge).

I do get annoyed at many of the legacy players putting out cars with good "range" but aren't that efficient, & are too expensive for mass market partly because they have oversized batteries. Not necessarily the case here but as I think you know most of the legacy automakers have been doing the wrong things with BEVs for years now. Tesla makes choices that are often not top of the line in order to keep costs down, but still manage to have good specs.

Ford's skunkworks project is more than 5 years behind Tesla. Meanwhile Tesla is already moving on & working toward providing transportation as a service. If successful (cost / quality / convenience) it will disrupt their own core consumer car business as well as the whole transportation industry. Legacy automakers still can't make, don't want to make, a true mass market BEV. Some options seem aimed at mass market but their sales numbers prove otherwise.

Anyway, am I mad? A little. I'm definitely annoyed & frustrated & sometimes discouraged at the US falling behind China which is making / has made the cities of the future while we are still struggling to get out of the 20th century because of legacy drag (& burning trillions on endless wars!).

This car is probably great but as I said it does nothing to move the needle. As none of the legacy options really have yet for a lot of reasons, mostly because they don't want to move the needle. They want to keep the status quo for as long as they can.

[Andrew Favreau @amf207]

I want to see the results of a simulation of this incident running on HW4 14.3.x. Would it have reacted sooner & what would it have done?

Not much you can do when a deer is determined to run full speed into the side of your car but... there may have been enough time to brake & allow it to pass in front if noticed quickly enough. 🤷