Comments by "Joe Qi" (@i6power30) on "The future of the automotive industry is not FSD or Tesla, its Ford Pro" video.

@HAIDARAVEN And I'm surprised how many people still beleive in Elon Musk's next year promises? Who is more clueless Elon or them?
4
There is no FSD . which Tesla car can "full self drive"? Naming this feature FSD is borderline fraud, and blatant misleading marketting.
2
@nickmcconnell1291 humans are flexible and adaptive but not infallible.. a small number of humans were fooled by the marketing term FSD and have already died in firey crashes unfortunately. Elon musk has long voiced his concerns over the harm AI could bring to humans. Yet he is doing just that.
2
Jim Farley is right. FSD is fool's errands. Better to focus on their strength. AGI is still 20 years away
2
@nickmcconnell1291 Nah, I already knew video generation was possible even 5 years ago, it was just matter of waiting for faster hardware and packaging it nicely for consumers like Open AI did. There is nothing revolutionary here. A pocket calculator can do multiplication millions times faster than humans, doesn't mean it can reason. AI machine learning can play chess better, can generate videos based on pattern matching, doesn't mean it can reason. Humans can learn from a handful examples, machine learning models need millions. They lack flexibility and general reasoning of human brain. In a game simulation, maching learning algorithm will be eatn by a lion a million times in various different ways before it learns to avoid it every time, not just sometimes. Even the dumbest human players will learn to avoid lions EVERY time, after just being eaten once, or twice. That's why sometimes you get werid mistakes from AI, such as a realistic looking picture of a farmer, but he has 6 fingers in one hand. Or videos with 3 wolves emerging out of just one body. Sometimes in Tesla FSD, a glare or smoke, or flash or light will make it mistaken for sometime else, and brake hard. It's all just examples of werid errors even the dumbest humans will not make. They cannot generalize and reason in broader context. All we are doing now with the current AI is improving task-specific level, making them more accurate and faster, but will never be able to adapt and be flexible like humans. What if FSD car encounters a police officer redirecting traffic with hand gestures? How many different hand gesture data examples does it need to be trained on before it can accurately and reliably handle all situations? Not all police officers do gestures the same way, not all of them wear the same uniform, some have gloves on, some don't, some use 1 finger, some uses 3 fingers etc. etc. What if there is a road rage, can AI read the "emotions" of the scene? Humans can sense danger and steer away from such situations. There many many such examples that the current AI machine learning bots simply unable to handle as well as humans can intuitively. I've been in this field many years. I hate to use the term AI because we don't have it yet. What we have is machine learning model of nueral network, which is not a model of human brain or mind, but simply a computer algorithm optimized for pattern reconition and reconstructure. It does not have ability to reason or understand which are both required for being called intelligent. You can ask a machine to generate a picture or a video of roomful of people at party. Then Ask the machine learning AI what is the vibe or mood of the roomful of people, and who is at the center of attention. Is he/she feeling happy or sad? It will not be able to answer these simple questions, but average humans can effortlessly.
2
@doittoit00 We still don't have FSD - nothing is "fully driving itself" yet. Yes, there were a lot of improvement, but still not good enough to fully replace human, and won't be for many years. It's easy to conquer the first 95%, it's the last 5% edge cases that's exponentially harder, and it's the last 0.1% that's almost impossible. Read my other reply, you will see what I mean. It's not just needing more data more compute etc. You need a fundamentally different architecture. You can throw all the money in the world at it (brute force machine learning), all you will achieve is smaller and smaller incremental imrovements, but fundamentally will not achieve AGI, by extension, no true level 4, 5 FSD.
2
Jim Farley is more intelligent than Elon Musk. Elon musk has said many things that were blatantly false.
1
@randalsaladbar I wonder why Andrej Karpathy quit. Maybe the world's best mind in AI realized the limitations of machine learning model and could not deliver what Elon asked of him
1
Ford is right not to waste money on FSD if it's not at their core competency. Look how much was wasted by GM's Cruise, Uber, Lyft, Google Waymo, Chinese Baidu, Nio, Tesla, all are racing towards FSD. None succeeded yet. The true AGI is still open for anyone to grab. It will take a whole new architecture that is drastically different from the current brute force machine learning to achieve it IMHO. It shouldn't take so much compute power and energy to train an algorithm a simple tasks. It shouldn't take miliions and billions of data points, it should take one a few dozen or hundreds of examples like training a human or monkey. Until then it might all just be fool's errand reaching for the impossible.
1
Where is AGI FSD? Who is developing them? AGI doesn't mean requiring massive compute and big data that currently Tesla is wasting money on. It's a dead end. If there were to be an AGI FSD in 5 - 10 years, we should have seen at least some publications on scientific journals by now.
1
@ISuperTed I agree. 20 years is more realistic even on a optimistic side. Could be even longer
1
@Nordlicht05 true. Whoever gets hold of AGI first will likely dominate the world. I wouldn't be surprised if China gets it first. They are limited by semiconductor ban right now but the whole country and CCP government are supporting it, compared to the US government is not taking it seriously at all. The front runner Open AI is a small company by comparison and few people working in it. China recently developed an AI chip 3000x faster than Nvidia a100 and uses analog and light instead of electrons. It's a game changer if they could scale it up. Tesla by comparison is relatively old school approach but Tesla has the most data. But like I said all that data is useless without the correct architecture I doubt Tesla has it otherwise they would have achieved FSD level 5 by now.
1
@nickmcconnell1291 You couldn't be more wrong. I've been in AI research for over 10 years, and it's amazing to see Elon Must has not a faintest grasp of fundamental difference between brute force machine learning and AGI. He is trying to apply something rudimentary and inefficient to a task that requires more flexibility and complexity in a completely different nature. You can have 1000000 trillion miles of driving data, and 100000 trillion times faster computer than 1000000 trillion NVDIA A100 GPU running in parallel training the neuro-network for 10000 trillion years, it will still not be as good as flexivility and intuition of a chimpanzee's brain. Just because the machine learning algorithm can generate realistic looking videos (which is specific task), it doesn't have flexibility of reasoning and adapbility. For example, Ask it to read emotion or vibe of the room filled with various people, it will have no clue. Human brains takes just a few exmaples (a dozen or so), and 1/10000 of energy to learn a new task. The best machine learning algorithm takes magnitude more than that. It screams bruteforce, and the real AI breakthrough doesn't come from faster chips, or more data, it will come from a more elegant architecture, and maybe years away even at academic levels. All these mad chase for FSD etc is fool's errand. I will bet 1 million dollars, FSD will not come in 2 year,s not in 10 years. Just have another analogy. Your pocket calculator can do multiplication a million times faster than you, does it mean it's smarter than you? Sure the machine learning took pocket calculator to the next level, it can do pattern recognition, and even reconstruct based on some inputs, but they are still TASK-SPECIFIC. It cannot do general learning like humans do, and and key is that it takes million times more data to learn from - meaning they cannot reason and generalize like humans. In a game, They have to be eaten by lions 100000 times in various ways before they learn to run from lions. Humans will learn it after just a couple of times even for the most stupid players.
1
We still don't have FSD - nothing is "fully driving itself" yet. Yes, there were a lot of improvement, but still not good enough to fully replace human, and won't be for many years. It's easy to conquer the first 95%, it's the last 5% edge cases that's exponentially harder, and it's the last 0.1% that's almost impossible. Read my other reply, you will see what I mean. It's not just needing more data more compute etc. You need a fundamentally different architecture. Elon doesn't seem to understand this limitation. Mark Zuckerberg does.
1
@doittoit00 maybe they are doing something complex different other then traditional machine learning/ neural network in secret I'm not aware of. But so far they seem to be even behind Open AI.
1
@TheEvilmooseofdoom any criticism on Tesla is met with this kind of personal attack. Not surprised. Time will tell. You keep holding TSLA bag
1
Sensible comment. Those who thinks FSD will fully replace human eyes are clueless. They bought into too much Elon's hype and lies, thus inflated TSLA stock. They will be bag holders for the next 10 years.
1