Hacker News Reader

Comment by wudangmonk | Hacker News Reader

"I think his point might be valid for gpt3.5 but has he tried 4.0? its on a completely different level, a game changer!."

This is pretty much what I have come to expect from any post that has the words "AI" attached to it.

As far as the tesla is concerned, its fun watching the screen to see what it identifies, and its cruise control works pretty well but the FSD should NEVER EVER be used. I test it every now and then against some freeway separators, and 5 years later its still trying to crash into them because somehow the road markers trump over the side collision detectors that are telling it there is a wall on that side.

View on HN

There were a couple releases that were bad when I first got access, but the last couple have been good for me.

It is excessively cautious, but that's better than the opposite and I take over whenever it could annoy other drivers.

It's excessively cautious until it drives you under a semi trailer and decapitates you.

Yeah but once you’re decapitated driving is no longer a problem for you.

Maybe don’t watch Harry Potter when you should be driving if you don’t want to cast Expecto Cerebellum.

That was autopilot, not FSD

I know someone who was rear-ended on the freeway because he had to hit the brakes hard and the driver behind didn’t react in time. Thankfully nobody was hurt. Just pointing out that “excessively cautious” can carry its own set of dangers, even if it’s ultimately safer.

Some of the scariest rides as a passenger were with drivers who were excessively cautious.

Because it comes across as a lack of confidence, and suddenly every minute is filled with second guessing every aspect of their driving, with the added fear of having no control over perceived mistakes in their driving.

It didn't "come across" like a lack of confidence, it was a lack of confidence. And the mistakes were not "perceived".

That's what I tried to tell some human drivers as well: driving excessively cautious (below the common cruising speed) endangers all parties (even when the common cruising speed is slightly above the speed limit ). Forcing other drivers to overtake or break is not a good idea.

It's funny how all the grifters have moved on. Okay, the cars still can't drive themselves, but now surely AI is a year away from being conscious!

Or, in a slightly less jaded take, the dreamers. Maybe not everyone excited by technology is trying to cheat you?

Both exist. I’ve seen a good number of folks who’ve gone from cryptocurrency to NFTs to AI in fairly quick succession, and I’m inclined to see that as grift.

Maybe just the majority, but I'd flip it as well and say a lot of people seeking investment are also pushing AI on the dumbest things. I've just stepped back from a company that was doing this claiming they are using AI but only in rewriting marketing blurbs or something along the lines of analyzing log files, but zero usage in the actual workflows/products of the company. Reeks of desperation, or scammy to be less generous.

Sure. But you can’t let the assholes distract from the miracles. That goes for everything in life. If you look at what AI does today and what it did 10 years ago, we are clearly living in the future dude. It’s better than I dreamed it could be, and I’m super excited to see what the non assholes do next.

We've been living in the future since the 1950s. We put men on the moon. We have the microwave oven. We have supercomputers small enough to fit in your pocket. We've mapped the genomes. We've almost had the year of the Linux desktop, we've almost had fusion in 10 years. Now we're really close to almost having AI for 10 years?

Living in the future is just such an odd comment to me. I get your enthusiastic about it, but what you consider futuristic is different than what I would consider.

The moment a dreamer starts charging you for promises that never materialize they become a grifter. (See: Tesla and FSD)

Very hard to find the authentic dreamers behind all the hubris these days. And by these days I mean the past 20 years...

Frankly, it is 1 script away. GPT-4 + a self-feeding script is conscious (obviously). Just not as conscious as a human being (but much more conscious than a sleeping human being).

Man, all these companies and engineers that wasted billions over a decade trying to crack self-driving cars are gonna have egg on their face after reading this

What do self-driving car have to do with the question of consciousness?

Forget self-driving cars, lets talk about self assembling automatons, I'm talking about ... TRANSFORMERS robots in disguise!

There's known value in autonomous driving.

Susan Blackmore observed that you can implement all the known functionality of human consciousness by answering the question 'Are you conscious?' with 'yes'.

We don't have any broad agreement that it does anything else, and we know that we do lots of things unconsciously despite our commonplace beliefs to the contrary.

aww, so three lines of BASIC is conscious.

Blackmore's argument is that people say they are conscious when they are awake and you ask them, but that's its only known function. Your reply appears to assume this is wrong. Do you have an argument to support it?

Beep boop I think therefore I am beep boop.

Interesting, so someone who has never said those words has therefore never been conscious, and we can treat them as we treat anything lacking consciousness. Hooray, we just solved all human problems: if someone has a problem, just turn them into soylent green! But only if they don't say those words!

> so someone who has never said those words has therefore never been conscious

That is not the argument.

People may well be conscious (no matter what they've said). What is the function of consciousness? Intriguingly, upon some reflection, we know of nothing that could not be done without consciousness [1]. One can seek food, mate, avoid danger, and so on without consciousness. One can speak and be intelligent without consciousness. So, why did evolution endow us with consciousness? Big puzzle.

[1] And now for the footnote/joke (which illustrates the profound point made in the paragraph above): The only thing we know of that one cannot do without consciousness is this: to truthfully answer "yes" to the question "are you conscious?".

Upon almost no reflection, you could make a long list of things that cannot be done without consciousness, or at least cannot be done yet and we have no good reason to believe could ever be done without consciousness. Can you understand anything without consciousness? Can you have an identity without consciousness? The argument you're presenting is that if physical mimicry is close enough to the original then it is the same thing, if we just assume that consciousness is nothing. It's saying that Frankenstein's experiment was largely a success.

I’ve driven a few Teslas lately and have been extremely impressed by the performance and comfort. But the auto steer and fsd preview are definitely underwhelming.

It can’t see a red light that is 120’ away to begin slowing down, nor does it consider that the navigation calls for a right turn at the light!

My theory is that the camera array needs to be larger and potentially much more data needs to be integrated into the decision making.

Even the fsd preview is scary when semi trucks seem to bounce left and right by about five feet per second while I pass them on the highway.

This is not a criticism of Tesla - the cars are a superb experience overall — just a note that the idea of self driving working on normal city streets or even suburban roads is pretty far fetched and will require significant enhancement to the tech that is in the 2023 vehicles.

Before long OpenAI will be able to implement FSD by trivially handing the video feed and API calls to GPTX and asking it to drive.

I think the Tesla AI team is just unserious about the problem, probably largely because of Musk being a poor leader for the project.

> My theory is that the camera array needs to be larger and potentially much more data needs to be integrated into the decision making.

Or they could just use radar/lidar like everyone else.

oh yeah

https://www.forbes.com/sites/bradtempleton/2023/03/24/cruise...

https://en.wikipedia.org/wiki/Death_of_Elaine_Herzberg

Human drivers dont need lidar and radar

Human drivers also don't need wheels. We can't pretend humans and computers are the same today so wishful thinking that "we do it so cars should be fine the same way" is the recipe for killing people every time the human fails to intervene and prevent it.

I haven't seen a clear proof yet that lidar/radar equipped vehicles are insanely safer than the current Teslas

I can’t believe it could be possible that less data is better than more data when it comes to self driving car sensors. Why not also include audio data and ir/uv data too?

I realized today that the Tesla system is noticeably slow to react to stimuli. It reminded me of how some gas station pumps emit a beep a second or so after you press a button when entering the zip code. Laggy, underpowered embedded systems.

I disagree on the cruise control, especially in Europe where the streets are narrower, it constantly thinks that it will crash and applies brakes. I sold my Tesla because of this

I try to avoid driving near a Tesla these days as I've seen so many do incredibly stupid things in front of me.

Disagree about FSD not being used. I've used FSD for thousands and thousands of miles without incident.

Have you tried FSD beta 11.3.x?

Reply on news.ycombinator.com