Autonomous Car Progress

[Bladerskb]

What if
- Huawei is better than Tesla in China AND
- Tesla is better than Huawei in US ?

While that's logical with high probability of being true.

The argument being put forth by you, @powertoold @ZeApelido and others is that nothing matters but the unparalleled supremacy of data. Its the end-all.. be-all, the alpha and the omega! If we accept this assertion as truth, it should be a resounding victory for Tesla - a "game, set, match" scenario like Elon said. Why? Because Tesla supposedly has billions of miles of data as you have stated and that no one is even close.

If we follow this line of reasoning to its logical conclusion, Tesla should outshine Huawei dramatically upon the unveiling of their system. Not just outperform, but in fact, utterly decimate Huawei's Autonomous Driving System. It should be a display of such overpowering dominance that it leaves no room for comparison or debate.

After all, Tesla has a massive fleet of vehicles in China, amounting to millions. This fact alone should translate to a remarkable advantage when it comes to acquiring training data. It's a simple extrapolation of the logical progression of your argument.

 

[KArnold]

nothing matters but the unparalleled supremacy of data. Its the end-all.. be-all, the alpha and the omega! I

But is the premise accurate? Some top-of-the-line talent is also needed. Remember, Betamax or cassettes to VHS or CD - you need the right talent to make it work.

Granted, if I had Tesla's data I'd be throwing all kinds of money at those guys. And maybe they do.

Point is only I'm not sure data alone is enough. But, it's a lot.

 

[powertoold]

Point is only I'm not sure data alone is enough. But, it's a lot.

All the current advances in NN usefulness are from large and diverse datasets: gpt4, midjourney, Google Palm, etc.

If people can find examples of impressive NN models with small datasets, please share.

 

[EVNow]

While that's logical with high probability of being true.

The argument being put forth by you, @powertoold @ZeApelido and others is that nothing matters but the unparalleled supremacy of data.

Where ?

I don’t think you know my position at all. This is the good old straw man.

I’m just saying it’s stupid to say X is better than Y - based on how they work in one place.

 

[Bladerskb]

All the current advances in NN usefulness are from large and diverse datasets: gpt4, midjourney, Google Palm, etc.

If people can find examples of impressive NN models with small datasets, please share.

That's not an advance. Making an NN model bigger is not advancement, its just scaling. The actual advancements / breakthroughs are the invention of NN architectures.
Transformers, NerF, GANs, Attention, Pointer, etc.

Maybe you should read this "leaked memo from Google Engineer:

"While our models still hold a slight edge in terms of quality, the gap is closing astonishingly quickly. Open-source models are faster, more customizable, more private, and pound-for-pound more capable. They are doing things with $100 and 13B params that we struggle with at $10M and 540B. And they are doing so in weeks, not months. This has profound implications for us:​

​

• We have no secret sauce. Our best hope is to learn from and collaborate with what others are doing outside Google. We should prioritize enabling 3P integrations.
• People will not pay for a restricted model when free, unrestricted alternatives are comparable in quality. We should consider where our value add really is.
• Giant models are slowing us down. In the long run, the best models are the ones
  which can be iterated upon quickly. We should make small variants more than an afterthought, now that we know what is possible in the <20B parameter regime."

Google "We Have No Moat, And Neither Does OpenAI"

Leaked Internal Google Document Claims Open Source AI Will Outcompete Google and OpenAI

www.semianalysis.com

Infact midjourney that you mentioned is exactly proof of that. This is a one man self founded tiny team with tiny data & compute compared to the gigantic companies with unlimited compute. Yet their Generative AI far surpasses the huge companies. Same with Stable Diffusion, just one man and his model are better and not just that, they are small and even runs on regular PC, phones ,etc.

 

[powertoold]

That's not an advance. Making an NN model bigger is not advancement, its just scaling. The actual advancements / breakthroughs are the invention of NN architectures.
Transformers, NerF, GANs, Attention, Pointer, etc.

Maybe you should read this "leaked memo from Google Engineer:

"While our models still hold a slight edge in terms of quality, the gap is closing astonishingly quickly. Open-source models are faster, more customizable, more private, and pound-for-pound more capable. They are doing things with $100 and 13B params that we struggle with at $10M and 540B. And they are doing so in weeks, not months. This has profound implications for us:​

​

• We have no secret sauce. Our best hope is to learn from and collaborate with what others are doing outside Google. We should prioritize enabling 3P integrations.
• People will not pay for a restricted model when free, unrestricted alternatives are comparable in quality. We should consider where our value add really is.
• Giant models are slowing us down. In the long run, the best models are the ones
  which can be iterated upon quickly. We should make small variants more than an afterthought, now that we know what is possible in the <20B parameter regime."

Google "We Have No Moat, And Neither Does OpenAI"

Leaked Internal Google Document Claims Open Source AI Will Outcompete Google and OpenAI

www.semianalysis.com

Infact midjourney that you mentioned is exactly proof of that. This is a one man self founded tiny team with tiny data & compute compared to the gigantic companies with unlimited compute. Yet their Generative AI far surpasses the huge companies. Same with Stable Diffusion, just one man and his model are better and not just that, they are small and even runs on regular PC, phones ,etc.

Nothing you said contradicts what I said though. I was talking about large and diverse (and accurate) datasets.

 

[Bladerskb]

Nothing you said contradicts what I said though. I was talking about large and diverse (and accurate) datasets.

Its literally the opposite of what you said in everyway.

 

[DanCar]

Below video mostly meant to sensationalize and get views.
Driverless cars creating traffic jams in San Francisco

 

[EVNow]

Below video mostly meant to sensationalize and get views.
Driverless cars creating traffic jams in San Francisco

How so ?

Everyday they are getting 3 calls to 911 of the AVs blocking traffic.

Sounds like a massive failure of safely pull over feature.

 

[Goose66]

Its literally the opposite of what you said in everyway.

Don't confuse the size of the model (or the size of the input vector) with the size of the training/validation/testing datasets. As Chollet said, if you have more time/money to spend, you probably much better off spending it in data gathering and curation than improving or increasing the size of your model.

 

[diplomat33]

https://twitter.com/x/status/1658048207879028737

 

[Goose66]

Below video mostly meant to sensationalize and get views.
Driverless cars creating traffic jams in San Francisco

Pure speculation and fear mongering, even requiring the reporter to add "and it might even pose a threat to safety, right?" What has happened to journalism in this Country?!?

 

[EVNow]

Pure speculation and fear mongering, even requiring the reporter to add "and it might even pose a threat to safety, right?" What has happened to journalism in this Country?!?

You are ignoring obvious safety issues.

Aren't L4 vehicles supposed to pull safely to the side instead of blocking fire-engines ?

 

[Goose66]

You are ignoring obvious safety issues.

Aren't L4 vehicles supposed to pull safely to the side instead of blocking fire-engines ?

I'm specifically addressing the lack of information in the report. The reporter says "she said she was getting three 911 calls a day regarding autonomous vehicles...." Why is he relying on what "she" said? 911 calls are public record, so go out and collect the statistics before you confront the operators. And also, why no statistics from the operators and/or what their mitigation capabilities are, what steps they normally take, and any changes they are going to make? It's hard for me to believe that the reporter doesn't understand that is exactly the questions all of the viewers are going to have - why put out a story without that information (or even acknowledging why that information is not included). It's just a bunch of people giving subjective opinions that it is or is not a problem instead of objective information regarding an actual problem - which is exactly what's wrong with our Media in the U.S. today.

 

[stopcrazypp]

I'm specifically addressing the lack of information in the report. The reporter says "she said she was getting three 911 calls a day regarding autonomous vehicles...." Why is he relying on what "she" said? 911 calls are public record, so go out and collect the statistics before you confront the operators. And also, why no statistics from the operators and/or what their mitigation capabilities are, what steps they normally take, and any changes they are going to make? It's hard for me to believe that the reporter doesn't understand that is exactly the questions all of the viewers are going to have - why put out a story without that information (or even acknowledging why that information is not included). It's just a bunch of people giving subjective opinions that it is or is not a problem instead of objective information regarding an actual problem - which is exactly what's wrong with our Media in the U.S. today.

Much of the news is subjective opinion and anecdotes instead of objective evaluations based on data. That's how news has worked since forever. Also news rarely waits for all the information to be available or the story would be stale by the time they post it. They try to balance things out by posting "both sides" most of the time.

 

[diplomat33]

https://twitter.com/x/status/1658182770869547009

 

[ThomasD]

Isn't the car programmed to backup at an intersection if no one is behind it to let a long vehicle like a Firetruck or Semi to proceed?

 

[diplomat33]

Great news for Cruise and Waymo. CPUC rejects the protest letter by the SFMTA and grants Cruise and Waymo permission to deploy driverless in all of SF, 24/7.

https://twitter.com/x/status/1658222810651557890

 

[ThomasD]

Is this for just Downtown or does this encompass areas outside of the Downtown City limits? Like the Airport

 

[Doggydogworld]

Is this for just Downtown or does this encompass areas outside of the Downtown City limits? Like the Airport

Airport is in San Mateo county. Waymo's application is for San Francisco and parts of San Mateo county. It also includes speeds up to 65 mph. So it sounds like they have airport service in mind. It could be a while, though. Maybe a long while. They still don't exceed 45 mph without a safety driver. And they don't serve Phoenix airport itself, just the SkyTrain terminal.