Future of Artificial Intelligence

[Henry]

The humanoid bots industry seems still in its nascency.

Seems or is? As technology advances the rapidly of the technology advancing accelerates. What used to take decades, now takes years. What used to take to take years, now takes months etc. They are not laying railroad track or building highways. Yes, ChatGTP doesn't spell check perfectly now, but one day it will and Candide's posts will actually make sense. Humanoid robots will be the largest TAM ever and there are already many players. TSLA is just the only one available to the retail investor. One day, everyone will wake up, and there will be FSD and bots. I'm investing like that day is today, because waiting for that day to happen and thinking it's some opaque future IMHO is neither wise nor the reality.

[7Wannabe5]

@Scott 2:

Very interesting article. If you combine that with the "rule" governing the increasing amount of compute available across spectrum of devices, in a few years, your garage door opener will have that much "independent" intelligence. Maybe I will go prompt Veo 3 with "scene in the style of Wes Craven featuring AI garage door opener 2027 in death battle with a Cybertruck attempting to convey child actress Drew Barrymore and her pet sea otter to their school pod."

[jacob]

Several shots have been fired for apparently no reason. Please take it to PM if you must continue, otherwise it ends now.

[suomalainen]

Autonomous driving is a revolution? [scratches head] fwiw, I think leon is more crazy than genius, and I would never get in a vehicle of his with no human with equal stakes in the trip’s outcome. I’ve ridden in an uber tesla once or twice, and while “neat”, it did the same thing the hondas and toyotas and fords do - drove me from point a to b. I don’t see any revolution in transportation happening until we get one of those beam ray transporter things.

In my neck of the woods (lawyering), AI is pretty pathetic. It’s not even as good as those past revolutions from word processing to email to redlining programs or even to, you guessed it, a form document. Will it get better? Sure. But until a client is willing to outsource judgment from a human professional to a machine, I don’t see AI replacing anything more than the most rote, brainless, first-year analyst tasks in law or finance or other professions that churn out similar documents time after time. Will machine judgment ever replace human judgment? Never say never, I guess, but that seems a leap I don’t really want to be around to see. Machines buying and selling stocks to each other in order to increase their own personal portfolios? [Shudder]

[Henry]

I would never get in a vehicle of his with no human with equal stakes in the trip’s outcome.

Have you ever been the sole passenger in a high rise elevator in the middle of the night?

[suomalainen]

Not one made by leon

Edit: two other points, one fair, one not: 1) i do get freaked out by elevators on occasion, but I understand the limited axes of movement and the non-existence of intersections (both human and machine) as well as the physical safety mechanisms built in should something over-accelerate. Roads and cars are different on each of these considerations. 2) I was a huge admirer of elon back in spacex days. Super cool stuff. Tesla was always a shrug for me. I’m not a car guy. But then it’s like leon lost his mind. I dunno if it was the money, the drugs, the self-hatred or what, but the guy could snap at any moment and I wouldn’t put it past a crazy fuck to have all of his cars drive off the cliff to join him on or before his own trip down the river styx.

Edit 2: they’ve had autpilot in planes for years. Decades? I still wouldn’t get in a plane that didn’t have human pilots in it. The max being the prime example of an “ai” that was programmed to kill the entire plane because it didn’t have human-level intelligence. I avoid flying in maxes even though the programming was “fixed”. I prefer forms of transportation that aren’t prone to self-immolation.

[Henry]

Well thank God we can trust every human driver on the road.

[ducknald_don]

https://en.m.wikipedia.org/wiki/List_of ... _Elon_Musk

If society really wanted autonomous vehicles then the best way of getting them would be to constrain the environment they operate in. Sort of like Heathrow Terminal 5 on steroids.

https://www.youtube.com/watch?v=BF1RVbnzPfs

[Jean]

We can at least punish human driver.
I find it very disturbing that self drivibg taxies can create accident without anyone being responsible.
There is always someone that decided to launch something with enough kinetic energy to kill several human in a public space.
And turning all street into a private space for cars only would be pretty dystopian.

[Henry]

Autonomous vehicles will be insured because people or corporations will own them but will cost less to insure because they will be safer by a wide margin.

As an FYI, the TSLA Model Y is the most popular car being sold worldwide, although not in the US. There are currently 5 million on the road. Both humans and AI use these stats.

[Jean]

Maybe they will, but if we have to make streets even more car centric to make it happen. That's not a win.
Do you think autonomous vehicles will also be safer on streets shared with pedestrians and cyclists?

[chenda]

One advantage would be we could easily enforce lower speed limits by executive fiat. This would make roads much safer for pedestrians and cyclists.

[Scott 2]

If society really wanted autonomous vehicles then the best way of getting them would be to constrain the environment they operate in.

This. Same for autonomous robots. Reduce degrees of freedom to make complexity manageable. The self driving problem benefits from acting in a largely 2 dimensional plane. General purpose humanoids face a challenge several orders of magnitude more difficult.

Scale discourages the generic solution. A robot chef would be cool. But a robot warehouse making meals, transported via delivery bot, heated on demand in an intelligent toaster oven? Far, far less expensive. Society is still refining that level.


@Henry - what car do you drive? If not a Tesla, how often do you ride in one? My few experiences as a passenger were underwhelming. It feels like I'm watching their competitive moat erode. See the New Jersey turnpike decision to replace Superchargers with Universal Open Access EV chargers.

People I know on the left, absolutely despise everything Elon. His legislative play may have caused irreparable damage to the brand.

Technology always commoditizes. The lifecycle is well understood, with tools to manage it. See this example of a Wardley map:



The competitive moat comes from broader factors. Tesla's had Elon's halo effect. I don't know how they restore it. The play into AI space with Grok and humanoid robot demos are interesting gambits. But I haven't seen them as differentiators. Tech alone isn't enough.

[jacob]

If society really wanted autonomous vehicles then the best way of getting them would be to constrain the environment they operate in. Sort of like Heathrow Terminal 5 on steroids.

... maybe a network of overhead wiring that the vehicles must follow at all times. This wiring could also deliver the energy and avoid the need for individual batteries in each vehicle. Indeed, such a network of vehicles could be coordinated from a central spot using time tables generated by AI!

Methinks the SV techbros need to get right on this. More ideas here: https://en.wikipedia.org/wiki/Trolleybus

Also see, https://stanforddaily.com/2018/04/09/wh ... -city-bus/

[candide]

The play into AI space with Grok and humanoid robot demos are interesting gambits. But I haven't seen them as differentiators. Tech alone isn't enough.

But even that creates bottlenecks and forces trade-offs. The training runs for the autonomous models still use Nvidia chips. Compute is still finite. It cannot both be used to push the autonomous training at the same time it is used to push Grok. (Yup, research with done GPT, but you can follow the links. I also even questioned it on a quote)...

The FSD chips that are vertically designed and implemented by Tesla are not some form of magic. They run the model, but they are not used to train the model. (Their engineering virtue is reduce latency). And the Dojo chip is not ready to work as an Nvidia replacement. Also, like so many key chips, both are made in Taiwan, which still should invoke certain um... risks.

[Henry]

@Henry - what car do you drive? If not a Tesla, how often do you ride in one? My few experiences as a passenger were underwhelming. It feels like I'm watching their competitive moat erode. See the New Jersey turnpike decision to replace Superchargers with Universal Open Access EV chargers.

We own two Subarus, one's a beater. I have to drive as it's part of my occupation. NJ is a blue state. The product will win. The economics will win. The insurance companies will want it. Once FSD is launched in geographic areas amenable to TSLA, such as Austin, it will have a domino effect and the chargers will be back. I live in an affluent area that is a red district. I run into Christine Todd Whitman in the library and say "Hey Gov." Well, actually "Hello Governor" but I'll get there with her. There's a lot of Teslas. Competitive moat? Who is the competition? No infrastructure needs to change. Don't need to move a stop sign an inch. It's simply a change in the auto industry - from ownership to service which we have seen begin with Uber/Lyft. So instead of some sketchy character pulling up in his dirty old ass car to drive you to the airport, a nice sleek Telsa shows up driving itself. Motherfuckers jump out of airplanes, paraglide and travel the world to ride rollercoasters that make me shit in my pants just by looking at the pictures. Now everyone's afraid to get in a car that drives itself? What are we? A country full of bitches? Grandmother's who were once afraid to pay their bills on the internet are now riding dildos the size of The Rock's forearm on porn sites. Or at least that's what I heard. It's a psychological transition that will have a network effect. The old people who shouldn't be driving will no longer be driving. Also, when people see that they can travel a mile for under $1, that's all she wrote.

The Elon factor will dissipate. People still use Facebook, buy iPhones, use Microsoft products. Zuckerberg, Jobs, Gates all went through it. Whatever you want to say about Elon, he doesn't make his workers keep their eyes open with clothespins like the manufacturers of all the other crap we buy in those big boxes of bullshit littering the place.

[candide]

I personally wouldn't mind getting into a hypothetical future Tesla robo-taxi as at a cheap price point (but I'd rather have a series of trolleys and/or trains that get me where I want in my metro, as that would have been at an even cheaper price point). But the second another service is cheaper, I would use that. This is called commoditization, and it is a different dynamic than a network effect. Your thesis seems to be that Telsa will be full robo-taxi when Musk says, even as he cannibalizes the compute he has for other purposes. And then, you are further contending that no company will ever figure out how to do this as well?

Zuckerberg, Gates, Jobs, all assholes, correct. But when you sieg heil you are going to lose some of your audience for good. Also, there are government workers who will remember the chainsaw. Not even just ones that lost their jobs. Example: I know some weather service dudes who would really like their grant money back; they just checked in one day and the account had been cleared out. They are not going to be buying a Tesla... Can't speak whether they would step on the hypothetical taxi. Again, it is a commodity, and supposedly free country, so there will be plenty of competition in time, if there is profit to be made.

[candide]

You know what, though? Same logic of commoditization should apply to Uber. It turns out people really don't like downloading a different app, and don't tend to do local apps.

Uber's market share is around 75% in U.S. Would Tesla getting 60% of rides (I mean some penalty for ruined livelihoods and offending some people) be a vindication of the "moonshoot" thesis, and even current valuation? Maybe. I'm still not changing any stock positions on this at present, but this seems reasonably possible. (But if they become Lyft...?)

One point here, though, is if Tesla is in the cheap robo-taxi business, wouldn't that take away from their car sales and benefits from their power station lock-ins? They might end up going from a high-margin business to a lower margin one.

[Henry]

One thing to consider is that TSLA investors are not limited to a select group of militant wack jobs posting on obscure websites who don't really have the slightest clue about what is going on but defend Elon Musk as though he truly rose from the grave driving a Cybertruck by regurgitating the thoughts of others which only exacerbates an already deeply ingrained anti-Elon Musk sentiment? I mean really, who wants to be that fucking asshole in the chat room?

Institutional money flowing into TSLA, especially since the latest pop, have greatly increased. Highly thought of investors such as Ron Baron are also out there front and center encouraging OPM investing. If history has proven anything, when Wall Street becomes the stakeholders, government responds accordingly. Sure they might hold a dog and pony congressional hearing, but at the end of the C-Span day, Mark Zuckerberg jet skis away.

[Scott 2]

We own two Subarus, one's a beater.

What stops you from owning a Tesla now? My neighbors aren't buying. They're moving to hybrid or electric solutions from traditional automakers. Or they have big ass trucks.

That's why I ask if you've ridden in a Tesla. It's fine, but not better. Comparable priced vehicles from traditional makers offer a more comfortable passenger experience.

The selling point was promise of the AI future. It didn't get delivered, competing north stars arose, then the figurehead alienated half his customer base. What's the network effect that will recover Tesla from that fumble? Facebook had billions of users, Microsoft every corporate desktop.

The American regions most likely to embrace self driving are technologically forward cities, like Austin. Where the brand is now toxic.


I think we have to be careful about conflating technological progress with a specific organization. Software (self driving algos) is free to copy. Faster progress buys a lead, but is never going to dig the moat. There has to be something more.

I do think we'll reach a point where human drivers are prohibited from the roads. My bet is on a timespan of decades though. I'm not seeing how Tesla protects a lead. Especially internationally. What stops a Chinese copy/paste into the budget EVs? How much time does that delay buy?

The world does not care about software IP rights. All of these AI efforts face the same risk. See the progress of running local LLMs, estimated to trail a few months behind the distributed commuting hotness. That's underwritten by leaks into the wild - meta's llama, China's deepseek. Models that cost billions to train are trivially downloaded.


First to superhuman AI? If achievable, that could be winner take all. It's what all dystopian authors warn us of. I personally think we're still in a phase of heavy disruption, with the victor entirely unclear.

Those controlling existing devices certainly seem to have the advantage though. Gemini instantly has an install base of every single android device. If data and compute are the keys, that's tremendous. Similar arguments for the other big tech firms.