I think self-driving targets a problem that doesn't really exist. The issue isn't that the act of driving is a laborious task, it's simply the amount of time spent in a car, which FSD doesn't address.
FSD isn't a complete product. Somehow they got away with selling an early beta for thousands of dollars. Zoox, despite an objectively odd priority on building a purpose built vehicle, became a generally available product ahead of FSD. That should be shameful.
Too much time spent inside may be a problem, but FSD turns car cabins into rooms. If we're inside already, a room with a destination is often better than a stationary one.
Now to start a tangent, what's the easier problem to solve: FSD, or a robust public transport system? Moving rooms have always been around in the form of trains, busses, streetcars etc...
Turns out, we have an answer to this: Self driving is easier. By a lot. It's not even close. No entrenched interests trying to block your infrastructure plans by claiming that your rail line passes through the territory of some flightless owl, no need to be called racist for cutting through the cheapeast land in the city to build out rails, no need to dig tunnels for subways. No need to overcome class prejudices where the middle class don't want to ride BART with the naked dude with a needle behind his ear.
To people outside the Bay, self driving might still seem like some far-off future tech. I can tell you that the future is already here. I haven't used an Uber in the last 3 years because I will always pick Waymo instead.
Fair enough. I will ask, how many billions have been spent in not only FSD but the car infrastructure that makes room for FSD investment?
I'm being slightly fanatical, but if our priorities were not car-centric in the 50's, do you think we would have spent more, or less money over the last 70 years on the transportation economy?
You're assigning decades of infrastructure costs to AVs, none of which was done with the intention of supporting them. They'd work fine on 1950s roads.
The global AV industry has taken roughly as much funding as California HSR has on its own, and less than what HSR will need to finish.
I've been doing public transit advocacy for my entire adult life. I've worked in the AV industry somewhat less than that. My advocacy has produced a couple of bike lanes and bus stops, contrasted against 3 AV launches.
I'd love to build more public transit, but experience tells me that the most effective thing I can do to support my community is AVs.
I do agree that it's not the panacea some people are hoping but true self-driving would change the experience for many people from a couple hours a day of not doing anything other than listen to music / podcasts / audiobooks to being able to do real work if they have things which can be done a laptop. Since multiple generations have been moving further out to car-only suburbs, I think that'd be very popular even if it's still not as nice as having a shorter commute.
It absolutely targets a problem that exists. Even in places with pretty great public transit, there is some demand for taxis/Uber/etc. Oftentimes even moreso, because if I don't need a car for 90% of trips, I might not have a car at all. So I use an Uber or a taxi when a certain trip demands it.
By far the worst part about said Ubers and Taxis is the driver. They're an unpredictable element in a situation where I greatly appreciate predictability. Unlike my parents, I didn't grow up with staff, so I'm not used to simply pretending this person I'm sharing a space with doesn't exist. Instead, I need to navigate the fuzzy line between courtesy and service.
Waymos have none of this shit. They're clean, show up when they say they will, I can play my own music, adjust the air conditioning, and have obnoxious conversations with my friends. They drive safely, and, as a cherry on top, they're cool as hell.
> It absolutely targets a problem that exists. Even in places with pretty great public transit, there is some demand for taxis/Uber/etc. Oftentimes even moreso, because if I don't need a car for 90% of trips, I might not have a car at all. So I use an Uber or a taxi when a certain trip demands it.
This says nothing about self driving cars
> So I'm not used to simply pretending this person I'm sharing a space with doesn't exist. Instead, I need to navigate the fuzzy line between courtesy and service.
I don't mean to be harsh, but, get over it? We live in a service economy. Do you feel the same way about the barista taking your coffee order?
> Waymos have none of this shit. They're clean, show up when they say they will, I can play my own music, adjust the air conditioning, and have obnoxious conversations with my friends. They drive safely, and, as a cherry on top, they're cool as hell.
I don't like the assumption you're making that Waymos are the only solution to ubers, taxis or driving yourself. Well designed and well working public transportation (Which is doable and exists in the world) is far cheaper and far more predictable than any form of car-based transportation.
Not only that, but you're not responding to my actual argument. The annoying part of driving is not the act of driving, it's the time spent in your commute.
> I don't like the assumption you're making that Waymos are the only solution to ubers, taxis or driving yourself. Well designed and well working public transportation (Which is doable and exists in the world) is far cheaper and far more predictable than any form of car-based transportation.
I very literally did not make that assumption. I pointed out, in a sentence you quoted yourself, that public transit can drastically reduce the amount of point-to-point personal transportation an individual wants or needs. However, sometimes, you really can't beat the convenience of "I am at point A, I want to be at point B, and I don't want to deal with a series of stops and transfers to get there". Maybe your starting or ending point is an unusual location. Maybe it's an unusual time of day. Maybe you're wearing a tuxedo or a cast and don't want to do the amount of walking public transit normally requires.
In any case, point-to-point transit is sometimes worth the expense. And when it is, self-driving taxis are fantastic. Compared to driving myself, I don't have to commit at least 75% of my attention to not killing myself or others. I can just read a book, or watch a movie, or do the morning crossword. Compared to taxis or Uber, I don't need to deal with a driver.
Point-to-point is also the only option when you get way out of the city and no form of public transportation is work-able. If you live in the actual middle of nowhere, with miles between homes' driveways, and dozens of miles between residential areas and the nearest store, you're never going to get trains or bus stops that cover everyone's home.
The route share option, which does sound like a minibus/dollar van, is interesting.
I've tried the current basic share option and it's not great, and I say that as someone who used pre-pandemic UberPool. You typically don't save much off a standard UberX ride, it's only available for exactly one person, the arrival estimates are wildly optimistic, and if the other rider isn't in the car they seem to never be ready when you get to their pickup location.
It's unfortunately, but the current pricing model seems to attract passengers who really don't want to be paying for an Uber but at least this way they can save a couple of bucks, which means they're typically in a stressful situation. Very different vibe from the old, social and wildly cheap UberPool, but that probably was never sustainable.
I feel like you're kinda proving too much. By the same reasoning, humans/programmers aren't generally intelligent either, because we can only mentally simulate relatively small state spaces of programs, and when my boss tells me to go build a tool, I'm not exactly writing raw x86 assembly. I didn't _build_ the tool, I just wrote text that instructed a compiler how to build the tool. Like the whole reason we invented SAT solvers is because we're not smart in that way. But I feel like you're trying to argue that LLMs at any scale gonna be less capable than an average person?
So far there hasn't been a transformative use case for LLMs besides the straightforward chat interface (Or some adjacent derivative). Cursor and IDE extensions are nice, but not something that generates billions in revenue.
This means there's two avenues:
1. Get a team of researchers to improve the quality of the models themselves to provide a _better_ chat interface
2. Get a lot of engineers to work LLMs into a useful product besides a chat interface.
I don't think that either of these options are going to pan out. For (1), the consumer market has been saturated. Laymen are already impressed enough by inference quality, there's little ground to be gained here besides a super AGI terminator Jarvis.
I think there's something to be had with agentic interfaces now and in the future, but they would need to have the same punching power to the public that GPT3 did when it came out to justify the billions in expenditure, which I don't think it will.
I think these companies might be able to break even if they can automate enough jobs, but... I'm not so sure.
Whatsapp had $10M revenue when it was acquired[1]. Lots of so called "chatgpt wrappers" has more revenue than that. While in hindsight Whatsapp acquisition at $19B seems no brainer, no concrete metric pointed to that compared to him investing $19B in AI now.
Dude Zuckerberg bought whatsapp because FB Messenger was losing market share... nothing to do with Whatsapps revenue! Rather Zuckerbergs fear of FB products being displaced.
How many software engineers are there in the world? How many are going to stop using it when model providers start increasing token cost on their APIs?
I could see the increased productivity of using Cursor indirectly generating a lot more value per engineer, but... I wouldn't put my money on it being worth it overall, and neither should investors chasing the Nvidia returns bag.
Why would the DoD give them money just to break-even (not even likely. oai has easily put > 10m into their compute) with companies that already have the infrastructure? This isn't a stimulus package, it's a service they're buying.
Even then, these companies aren't doing research into LLMs, they're just wrapping the endpoints and creating some abstractions.
Recently as in the last few days when it started calling itself "MechaHitler" and scapegoating jewish people after the engineers let Elon ramble for the system prompt.
I still crack up at the idea of 'personality prompting', mostly because the most engaging and delightful IRL persons who knock us off our guard in a non-threatening way are super natural and possess that "It Factor" that's impossible to articulate lol -- probably because it's multimodal with humans and voice/cadence/vocab/timing/delivery isn't 100% of the attraction.
That said, it's not like we have any better alternatives at the moment, but just something I think about when I try to digest a meaty personality prompt.
This character prompt has undergone so many iterations with LLMs it's not funny anymore. "Make her act more bold." - "She again talked about her character description, prevent that!"
I was hoping she'd let him have it for the way he kept interrupting her. But unfortunately it looks like he was just interrupting the TTS, so the LLM probably had no indication of the interuptions.
To me it's something like "the target language does not differ from the expressing language"?
A .txt file for notes is plaintext, because the language I'm using doesn't have to be compiled for my goal. Programming languages are not, because the expressed language is compiled into some other target language (machine code).
Markdown is not, because it's compiled into HTML.
A .txt undergoes no transformations from my writing, to its storage, to my later usage of it.
reply