Neural Networks

[heltok]

Was listening to Karpathys most recent talk:

He starts at 3:09.

At 11:45 and forward he says that he believes in "Building a single model". I assume that the bigger FSD NN that we have seen might is the single model he talks about. And I assume this is some common layers and then a lot of different layers for the different tasks. Something like A->(B,C,D,E->(F,G->H))) or something like this:

 

[strangecosmos]

A new Karpathy talk? Is it Christmas already?

 

[strangecosmos]

And no i don't think you should mislead a demographic by spreading false narrative and information to please what that demographic wants to hear. Jimmy's podcast is a tesla fan's wetest dream! (yes i know "wetest" is not a word but in this case it is!)

Propagating that Neural Network is a camera only position.
Propagating that only Tesla uses NN.
Propagating wrong info about the usage of NN.

That's some mighty reality distortion field. I absolutely think he did more harm than good.

This is such an absurd and offensive post. This the problem with Blader’s behaviour on this forum; why it is so disruptive and antithetical to real, meaningful technical discussion. Jimmy of course didn’t say that only Tesla uses neural networks. That would be ridiculous. Yet Blader wrongly claims that Jimmy is “propagating” this notion.

Likewise, Jimmy didn’t say that neural networks can only use camera input. That would of course be false. Blader seems to have persistent issues with reading/listening comprehension, or memory. That’s the generous interpretation.

Blader likes to claim that people said things they never said — typically false and ridiculous things. This is a recurring pattern. I feel like this warrants being banned from the forum. If you repeatedly spread insulting falsehoods about other members — when you definitely should know better — then you should be banned. This is worse than just calling someone stupid. At least then you are being transparent. When you make up insulting stuff about someone, you try to drag them down in the eyes of other people. You try to deceive other people as part of your effort to attack someone.

If you insult someone, it’s obvious that you are being obnoxious. If you make up an insulting lie, you can trick people into thinking you’re reacting reasonably, and that your target deserves reprobation. This is a far more sneaky and damaging tactic than just calling someone an idiot. It’s much worse. Much more disruptive.

If any mods read this post, please take this into consideration.

 

[heltok]

After watching that Karpathy video, I got inspired to brush up on what has been happing in the AI world. The guy leading OpenAI, where Karpathy came from and which Elon funded, had a pretty damned interesting video:

I think ideas from this will translate into Tesla Autopilot team.

 

[electronblue]

Here is an actual EyeQ4 visualization, taken from the BMW Emergency Stop video. It's not pretty. There is a ghost car sitting on the rear bumper, and the overtaking car disappears and fragments as it passes:

Later, we see what happens then the BMW starts changing lanes and is passed by another car that deforms and loses position as it passes. Again not pretty, but at least the tailgater has gone now...

If this was any other vendor, we would say that the vision/radar fusion is a bit rubbish, but because EyeQ4 is L5-ready, the only possible explaination is that the overtaking car in the first GIF is actually falling to pieces.

This is because of the micro-wormholes which inconveniently materialised on the BMW test tack as they were filming - the car disappears from the display every time it hits one. The second GIF shows the stretch effect of driving close to a micro-wormhole, without actually hitting it. Luckily that car didn't disintegrate too.

This is not EyeQ4 at all. This is an old video from 2010.

Please correct your post.

This original video is here:

The video is posted in November 7, 2010.

 

[J1mbo]

This is not EyeQ4 at all. This is an old video from 2010.

Please correct your post.

This original video is here:

The video is posted in November 7, 2010.

How do you know that the car in that video was not running a development version of EyeQ4?

The actual chip was launched in March 2015, planned for market in 2018. The feature will become available with EyeQ4-equipped cars. Seems highly likely to me!

 

[electronblue]

How do you know that the car in that video was not running a development version of EyeQ4?

The actual chip was launched in March 2015, planned for market in 2018. The feature will become available with EyeQ4-equipped cars. Seems highly likely to me!

Where did you get the idea that a BMW video from 2010 would be running EyeQ4? Engineering samples of the chip became available in fourth quarter of 2015, full five years later than the video was posted on YouTube. Five years is an eternity no?

Also the video seems to show side radars or lidar not a camera suite towards the sides.

 

[J1mbo]

Where did you get the idea that a BMW video from 2010 would be running EyeQ4? Engineering samples of the chip became available in fourth quarter of 2015, full five years later than the video was posted on YouTube. Five years is an eternity no?

Also the video seems to show side radars or lidar not a camera suite towards the sides.

Lol, where did I say engineering samples? The car probably had a trunk full of prototype kit.

Where is the LIDAR sensor in that video? In 2010 a LIDAR unit wasn't exactly discrete.

 

[electronblue]

Lol, where did I say engineering samples? The car probably had a trunk full of prototype kit.

Where is the LIDAR sensor in that video? In 2010 a LIDAR unit wasn't exactly discrete.

Can you give me some guidance on why you think a BMW video from 2010, several years before EyeQ4 launch, has an EyeQ4 in it or would be representative of EyeQ4 performance that launched half decade later and appear in production cars almost a decade later? I must admit I don’t follow your logic. Do you have a source?

Ibeo lidar can be seen at 1:02. Side sensors are at 0:50 that look like radars perhaps not cameras.

 

[conman]

Is it a neural network optimisation that they've done to only look for and assume vehicles travel in one direction or just a cluster related phenomenon? Whenever my V9 recognises vehicles on the opposite side of the road, they briefly look like they're going forwards before disappearing. Yes I understand the display does not match the NN output used for autopilot but they still share inputs at some level.

 

[brendon1970]

Is it a neural network optimisation that they've done to only look for and assume vehicles travel in one direction or just a cluster related phenomenon? Whenever my V9 recognises vehicles on the opposite side of the road, they briefly look like they're going forwards before disappearing. Yes I understand the display does not match the NN output used for autopilot but they still share inputs at some level.

Yours to huh. i wonder if the US cars do the same for oncoming traffic?

 

[conman]

Yours too huh. I wonder if the US cars do the same for oncoming traffic?

Haha I can't imagine they'd make such a big mistake on converting it for right hand drive, right? RIGHT!?

 

[Vitold]

Do we have any idea what method is Tesla using for 3D plane detection?
We know that current approach does not like shadows.

I found this paper quite interesting:

https://faculty.ist.psu.edu/zzhou/paper/ECCV18-plane.pdf

 

[Bladerskb]

This is such an absurd and offensive post. This the problem with Blader’s behaviour on this forum; why it is so disruptive and antithetical to real, meaningful technical discussion. Jimmy of course didn’t say that only Tesla uses neural networks. That would be ridiculous. Yet Blader wrongly claims that Jimmy is “propagating” this notion.

Likewise, Jimmy didn’t say that neural networks can only use camera input. That would of course be false. Blader seems to have persistent issues with reading/listening comprehension, or memory. That’s the generous interpretation.

Blader likes to claim that people said things they never said — typically false and ridiculous things. This is a recurring pattern. I feel like this warrants being banned from the forum. If you repeatedly spread insulting falsehoods about other members — when you definitely should know better — then you should be banned. This is worse than just calling someone stupid. At least then you are being transparent. When you make up insulting stuff about someone, you try to drag them down in the eyes of other people. You try to deceive other people as part of your effort to attack someone.

If you insult someone, it’s obvious that you are being obnoxious. If you make up an insulting lie, you can trick people into thinking you’re reacting reasonably, and that your target deserves reprobation. This is a far more sneaky and damaging tactic than just calling someone an idiot. It’s much worse. Much more disruptive.

If any mods read this post, please take this into consideration.

Lies? Everything statement I have made quoting you have been completely true.

If you want to know who @strangecosmos is. He is Trent from seeking alpha.

All you have to do is look up his 53 articles he wrote for seeking alpha going all the way back to 2017. Look at his statements. Trent tries to hide these facts that hes been wrong time and time again. But it's there for the world to see.

You can call ms a liar all you want but the truth of every statement you have made and continue to make is out there.

Trent Eady's Articles | Seeking Alpha


Secondly about jimmy. I have quoted exactly the statements jimmy made. He absolutely gave the impression that only Tesla used NN and that NN was a camera position. All you have to do is look at the quotes.


Not only that. Look at the commentaries done based on his pod cast.

Deep Dive Into Tesla's Autopilot & Self-Driving Architecture vs Lidar-Based Systems | CleanTechnica

Now tell me. Why did that article receive major blowback not only from the site fanbase but from teslamotors subreddit of all places?

Deep Dive Into Tesla’s Autopilot & Self-Driving Architecture vs Lidar-Based Systems : teslamotors

This is exactly what i predicted, that his podcast will lead to more myths. I was right yet again... Surprised?

Trent wants me gone so he can peddle his false narrative that have been proven wrong over and over again unfettered!

Like this one "Tesla Leapfrogs Self-Driving Competitors With Radar That's Better Than Lidar"

Trent pushed the narrative that Telsa narrow, non heated singular radar had superseded lidar because Elon made a tweet. Yet there have been dozens of accidents of AP running into cars since then.

Its countless the amount of whole-fully wrong articles Trent or rather @strangecosmos has written.

Trent considers the only experts in self driving to be elon musk and karpathy and anyone who agrees with him.

To give you a snapshot at how trent reasons when it comes to tesla.

Trent did a back of the envelope math and said that it only takes Tesla image taggers 5 seconds to label a picture. I called him out on how ridiculous that statement was and that it proves he hasn't done his research at all.

I posted videos about image labeling companies and tools and gave him a challenge to use one of these online labeling tools and label road images in 5 seconds. In conclusion I explained how it takes way more than 5 seconds to even look at an image.

And just a research of a grain of salt is required to come to that conclusion, therefore he is simply here to push the Tesla narrative.

Trent responded back and lashed out and attacked me and did not adjust his position/statements.

Weeks later, karpathy came out with a talk showing how hard it is for their labelers to label a single image.

After Trent then changed his position ONLY after watching the video and now calculates it takes the taggers 10 mins per image.

But I explained this to trent WEEKS AGO! What does this prove? Trent will only listen to ANYTHING that comes from TESLA and will disregard even cold hard evidence.


That's not a discourse that's a echo chamber! Ask yourself, do you want to have a discussion on neural networks or do you want to have an echo chamber with Trent leading the charge?


Tell me now and I will leave and never come back so you can listen to the guy who wrote dozens and dozens of articles arguing furiously that Tesla will have FSD Level 5 in a little over a month to a bunch of gullible investors!

 

[Vitold]

Lies? Everything statement I have made quoting you have been completely true.

If you want to know who @strangecosmos is Trent from seeking alpha.

All you have to do is look up his 40 articles he wrote for seeking alpha going all the way back to 2017. Look at his statements. Trent tries to hide these facts that hes been wrong time and time again. But it's there for the world to see.

You can call ms a liar all you want but the truth of every statement you have made and continue to make is out there.

https://www.google.com/url?sa=t&sou...gwAHoECAYQAQ&usg=AOvVaw2dA7xcrp7GqQsYPh2jWpDU


Secondly about jimmy. I have quoted exactly the statements jimmy made. He absolutely gave the impression that only Tesla used NN and that NN was a camera position. All you have to do is look at the quotes.


Not only that. Look at the commentaries done based on his pod cast.

Deep Dive Into Tesla's Autopilot & Self-Driving Architecture vs Lidar-Based Systems | CleanTechnica

Now tell me. Why did that article receive major blowback not only from the site fanbase but from teslamotors subreddit of all places?

Trent wants me gone so he can peddle his false narrative that have been proven wrong over and over again unfettered!

Like this one "Tesla Leapfrogs Self-Driving Competitors With Radar That's Better Than Lidar"

Trent considers the only experts in self driving to be elon musk and karpathy and anyone who agrees with him.

To give you a snapshot at how trent reasons when it comes to tesla.

Trent did a back of the envelope math and said that it only took Tesla labelers 5 seconds to label a picture. I called him out on how ridiculous that statement island that it proves he hasn't done his research at all. I posted videos about image labeling companies and tools and gave him a challenge to use one of these online labeling tools and label images in 5 seconds. In conclusion I explained how it takes way more than 5 seconds to even look at an image.

And just research of a grain of a tiny seed is required to come to that conlusion.

Trent responded back and lashed out and attacked me and did not adjust his positoon/statements.

A Month ago, karpathy came out with a talk showing how hard it is for their labelers to label a single image.

After Trent then changed his position ONLY after watching the video and now calculates it takes the taggers 10 mins per image.

But I explained this to trent WEEKS AGO! What does this prove? Trent will only listen to ANYTHING that comes from TESLA and will disregard to contrary evidence.


That's not a discourse that's a echo chamber! Ask yourself, do you want to have a discussion on neural networks or do you want to have an echo chamber with Trent leading the charge?


Tell me now and I will leave and never come back so you can listen to the guy who wrote dozens and dozens of articles arguing furiously that Tesla will have FSD Level 5 in a little over a month.

Based on Karpathys presentations Tesla is replacing their current AP code with NN heavy development. This is a huge departure from Mmbly and in-house AP code 1.0, therefore anything Tesla said in the past about AP is likely inaccurate as a result.

 

[strangecosmos]

Tell me now and I will leave and never come back so you can listen to the guy who wrote dozens and dozens of articles arguing furiously that Tesla will have FSD Level 5 in a little over a month to a bunch of gullible investors!

This is either 1) a bald-faced lie or 2) you really struggle to understand/remember the things you have read. You do make positive contributions to this forum, but blatantly false statements like this are disruptive and trollish. They outweigh the value you contribute.

I think you should find and link to an exact quote where I said I believed Tesla would have full, Level 5 autonomy by the end of 2018. If you can’t do that, I think you should admit your mistake, apologize, and be a lot more careful about misattributing statements to people in the future. Maybe you just genuinely misremember what people have said. If that’s the case, then you should double check your sources before posting and use exact quotes rather than paraphrasing. You have less risk of getting something wrong and offending people that way.

If you’re unable or unwilling to either cite an exact quote or apologize and try to be more careful in the future, then yeah, maybe it would be best if you left the forum. If you can’t stop yourself from muddying the waters by claiming that people said things they never said, then your participation here is probably unhealthy for you and for the people you are either advertently or inadvertently trolling.

 

[Bladerskb]

This is either 1) a bald-faced lie or 2) you really struggle to understand/remember the things you have read. You do make positive contributions to this forum, but blatantly false statements like this are disruptive and trollish. They outweigh the value you contribute.

I think you should find and link to an exact quote where I said I believed Tesla would have full, Level 5 autonomy by the end of 2018. If you can’t do that, I think you should admit your mistake, apologize, and be a lot more careful about misattributing statements to people in the future. Maybe you just genuinely misremember what people have said. If that’s the case, then you should double check your sources before posting and use exact quotes rather than paraphrasing. You have less risk of getting something wrong and offending people that way.

If you’re unable or unwilling to either cite an exact quote or apologize and try to be more careful in the future, then yeah, maybe it would be best if you left the forum. If you can’t stop yourself from muddying the waters by claiming that people said things they never said, then your participation here is probably unhealthy for you and for the people you are either advertently or inadvertently trolling.

When did i ever say the end of 2018?

You have been peddling the 2019 narrative for years. When is 2019? Wait...its a lil over a month from now. Its called Jan 1st 2019. Hate to break it to you 2019 doesn't start in June.

"Even if a competing car company wanted to change plans and race Tesla to launch self-driving in 2019, incumbent car companies have an incredibly slow cadence of hardware updates. It seems unlikely they could pull up their production timelines by the two to three years that would be required."


Whereas Tesla is a software company, incumbent car companies have no proven competence in software. It may turn out that even the self-driving systems slated for 2020 and 2021 are just vapourware."

"The earliest that incumbent automakers have said they plan to release a car with full self-driving hardware is 2020. Many are targeting 2021."

"During this window of one or two years, Tesla would experience virtually unlimited demand and a massive influx of capital, enabling the aggressive acceleration of its production ramp. With the launch of Tesla’s autonomous ride-hailing service, the Tesla Network, Tesla’s cars would drive 5 to 10 times more miles, i.e. as much as a human-driven fleet 5 to 10 times its size. Since these would be fully autonomous miles"

Not only have you been pushing 2019. But whats even funnier is how you misrepresent the market. The only thing automakers are planning to release are level 4 highway self driving cars. which is completely different from the level 5 sdc with a taxi fleet network that you continue to push that Tesla will have in 2019. You also added that you see a scenario where Tesla will completely stop selling model 3 at the start of 2020, after months of seeing their FSD tesla network adopted in 2019 and go full taxi fleet.

Then capped it off by saying what the trad automakers will come out with are actually vaporware but Tesla FSD are the real deal.

Its laughable...

And i found this info by looking at just 2-3 articles because all the old articles are paywalled and i don't plan on spending all day trying to defeat the system.

 

[Tjhappel]

Did one of you guys cheat on each other?? Trent Change your avatar picture please...

Keep up the awesome drama guys, it’s great reading and currently I’m more sympathetic and leaning towards bladder but I love jimmy’s info so.....

I’m confused

 

[Off Shore]

This was an interesting thread until it was taken over by trolls. The people who actually know something have gone silent.

Edit: ...as was the goal of the trolls.

 

[scaesare]

So you accuse @strangecosmos of stating:

...Tesla will have FSD Level 5 in a little over a month...

And, you actually provide the quote yourself wherein he actually said:

...launch self-driving in 2019...

(notice the emphasized word there.. that's key)

You justify that by expanding:

When is 2019? Wait...its a lil over a month from now. Its called Jan 1st 201

So while 1/1 is in 2019, so is 12/31. So they could introduce self driving in a little over a month, and he'd be right. Or they could introduce it in 13+ months... and he'd still be right.

But of course, that's not how you chose to represent his statement. And yetyou accuse others of lies and misleading claims...

Interesting.