WEBFI-TECH-What to Expect from Tesla FSD in 2022 w/ James Douma #24 (Ep. 482)
I offered for all of us to chip in to pay for James Douma’s new MacBook Pro, but he’d rather people give to the Afghanistan food relief project, https://www.gofundme.com/f/food-relief-project-by-dave-lee . Campaign will end on Jan 2 midnight CT. Big thanks to James Douma who donated $20k after the interview.
James Douma on Twitter, https://twitter.com/jamesdouma
James Douma on AI playlist, https://www.youtube.com/playlist?list=PLfibpgBinf9R7KIedEU3y-YjrA63LSKHX
Timestamps:
00:00 – Introduction
02:00 – FSD Beta 11 one stack to rule them all
05:24 – Surround video vs vector space
11:00 – Bag of points explanation
22:00 – Elon Musk on AI
25:15 – Raw photons vs image processing
33:00 – C Compiler explanation
35:00 – FSD timeline, Level 4 by end of 2022?
44:00 – James Douma’s personal experience with FSD Beta
47:46 – Tesla Dojo progress
51:17 – Tesla Bot expectations
55:30 – Tesla 2022 – what to look for
59:40 – James Douma psychology of TSLA stock price
1:04:15 – James Douma’s new MacBook Pro
1:10:23 – Dave offers for him and fans to pay for MacBook Pro
1:12:00 – Afghanistan food relief project
1:15:10 – Conclusion
Social
🐦 Twitter https://mobile.twitter.com/heydave7
🎧 Apple Podcast: https://podcasts.apple.com/us/podcast/dave-lee-on-investing/id1493031991
🎙️Spotify https://open.spotify.com/show/2iin015s0k21Bt4ldep9XT?si=uBV4Ft8hQEu_GE4YSIDDGg
📸 Instagram https://www.instagram.com/heydave7
Dave Lee on Parenting, https://www.youtube.com/channel/UCWD9b1BZqi2G2u30sGYIFrA
NOTE: Please don’t fall for scammers who might impersonate me or others in the comment section. I do NOT give out my number or ask people to contact me.
💁🏻♂️ Disclaimer: All content on this channel is for discussion and illustrative purposes only and should not be construed as professional financial advice or recommendation to buy or sell any securities. Should you need such advice, consult a licensed financial or tax advisor. All views expressed are personal opinion as of date of recording and are subject to change without responsibility to update views. No guarantee is given regarding the accuracy of information on this channel. Neither host or guests can be held responsible for any direct or incidental loss incurred by applying any of the information offered. Author is long TSLA, LMND, SQ, and other stocks at time of original video publish date. When you buy through links in this video description, author may earn an affiliate commission.
Open an brokerage account at Interactive Brokers:
https://www.interactivebrokers.com/mkt/?src=daveleey&url=%2Fen%2Findex.php%3Ff%3D1338
Buying a Tesla?
🚗 Car: Use my referral link to receive 1,000 free Supercharger miles with the purchase of a new Tesla. http://ts.la/david7556
☀️ Solar: You’ll get $100, I’ll get $500 for a new install of a Tesla solar system. http://ts.la/david7556
Check out my archived articles/posts on Tesla and investing: https://teslamotorsclub.com/tmc/threads/articles-megaposts-by-davet.23473/#post-485768
source
Stock compilers normally do low quality job on generating codes for modern processors: Mainly in area of loop unrolling, and vector instructions. For example Intel has its own c compiler "icc" which out perform gcc and clang. So I suspect Tesla is patching the current compiler to generate a better code. Which is very common practice for hardware teams.
It seems to me that FSD has a lot of problems navigating badly designed intersections. Maybe it would be better if the car simply avoided these intersections at the cost of taking more time to get o your destination. Or alternatively, we just fixed the bad intersections for the benefit of everyone.
If someone owns a Tesla with FSD and drives it until it gets to level 4, does a different driver have the same level of FSD or do they have to start over?
Investors want SINGLE holding company for THEIR advantage. not for Elon management of these companies right ??.
Larger simpler batteries for utility companies probably cheaper than small cells for vehicles.
Dave you have an eye for interesting guests. You put James, Emmett and Matt on the map, as they deserve. These guys, and your questions, are terrific.
It's baffling why this (or any James Douma) video doesn't have 1M likes…
Even if regulation allows for no human oversight of a self-driving car, being twice as safe as the average human driver is still untenable from a product standpoint. Whenever there is disruption in an industry, there is a general rule that requires ~10x improvement of the closest competitor in order to gain traction and reach a critical mass of users. In this case a need for the self-driving car to be ~10x safer than a human driver. I still believe we are at least 4 years away from having a car drive without a driver present. Much work to do.
This is not like Netflix disrupting Blockbuster by providing 10x more value. Getting people on board with self-driving cars is literally getting people to trust their lives with a product. IMO It will require a minimum 10x improvement over the average human driver and will gradually be accepted over time as people monitor the car's performance and trust it more and more day by day. Or perhaps even not trust it for some.
If he's had significant money in Tesla for a long time then he has enough money 🙂
It's interesting that he says Tesla has a guy coding an explicit driving policy, I don't think that's true for anything other than some special cases unless he's really bad at it 🙂 because FSD routinely drives into the path of other vehicles and that would be the simplest thing to code. It clearly seems to me that the driving policy is mainly programming2.0 learned.
Just a few dozen Tesla humanoid robots: super useful for 1st unmanned Star Ships to Mars/Moon. Very first ship lands on Unprepared surfaces; Sadly ship keels over, but Robots get out of crash, set up coms & cameras, and start shoveling a flat space… Set up useful equipment that survived crash. No human died. That first landing & crash ac , , shove
continued: first landing even with a partial failure uses robots to prepare site for next unmanned landing, prepare for a better and more stable landing site, etc.
We love spacex & we just got our S dual motor fsd while the tri/4? motor fsd cybertruck gets here. i'm all in TSLA & i love all Elon Musk does & will keep doing..we've drove from California to Boca Chica to see the Star Ship & it was so awesome to see it getting built in person!! we have LOTS of SOLAR & backup battery system to charge them both
Where can I get that FSD T shirt?
Douma is the best. I think I've watched every one of them on this channel. Head and shoulders above the second best guest.
Holy smokes:
Tesla is doing survey lidar without lidar. #Elonmusk this is another revenue stream. You could sell this to engineering firms to create existing topographic surveys.
If it takes top tier minds years to solve the problem it will take 2nd and third rate minds decades.
Why not have 2-3 of drones that would station keep, flying above the car, weather permitting, they would swap jobs as their batteries run low, land while driving,,to recharge while another takes off, feeding real overhead live video to the AI, to help it not miss anything from any direction, speed velocity info would be great and so would object avoidance! If a car was heading to T Bone the Tesla, the AI could pick a path and drive out of the way even left of center since it knows nothing is immediately coming down the road, , change into another lane and accelerate out of the path, or reverse out of the path because it knows it is not going to cause a rear end collision! The advantages would be incredible. This would simplify AI Decision Making greatly!
44:03 Red pill moment
C Code is rule based and very inefficient. However, they need to replace the C code not with classical ANNs but ontological high abstract symbolic ones….vectors are associative information with probability degrees. Tesla will have another major improvement ahead even after stack integration and the first level of C code replacements.
XLnt !
It's seems like they are still building up the levels of abstraction, from photons to bags of points to objects to behaviours in 3D space, gradually replacing a lot that is still done by the procedural C code. This makes sense but is further behind than you'd think from the latest FSD abilities. It's like it's just got past the peekaboo toddler stage in recognising object permanence, which is no longer procedural in being a objects in memory. It's on the right track though and learning fast. Modeling and predicting finer grained behaviours of 3D objects, especially ones that change shape or are clumps like piles of leaves or snow would be the next step that I can see being tricky but one that ML would handle much better than procedural code. One that's there there's refining the function from understanding the surroundings to the car responding correctly which will be a whole new set of abstractions for various scenarios.
You the man Dave, J.D. also. You guys are quite refreshing.
The guy on the right looks like Reviewbrah in about 10 years
Why won't you guys talk about FSD competitors one day?
I really love your channel I don't agree with everything everyone says but I really enjoy your content.
Would be interested in his take on Mobileye's update at CES and how it stacks up to Tesla
Hey, David. I have a question: Where I can buy the kind of T-shirt that James is wearing in your interview?
Accidents can't be ruled out totally impossible ever. We have wild animals running crossroad in panic. Those animals may weight some 1000 pounds and have some significant speed. Even if you can avoid hitting those animals that won't prevent those animals hitting you. Just a thought.
Dave Lee shared an amazing interview with James Douma. It gave me great insight how TESLA is developing FSD. Their processing of images into video and converting that to a 3D view in beyond my comprehension. Amazing. Thanks for sharing. Happy New year to you both. Looking forward to your next interview with James.
oh wait, how to train the robot – I think they can create an arm, and a head with eyes, and you can play catch with it. It would have to learn how to grasp, how to track a ball in the air and coordinate the hand in space to intercept the ball. To catch it will learn to decelerate the ball so that it doesn't bounce off the hand, and it will have to learn the nuances of throwing the ball back at different velocities and different angles. And this would be on their AI stack. Simultaneously another team could be teaching the lower torso how to balance standing bipedal and then give it acceleration, swiftness and jumping. When they put this together then the eyes can track the environment and then it can navigate the world at its maximum 5mph without knocking things down. At this point it should be able to juggle 3 balls standing on one foot, all self-learning. And now it can deliver telegrams, ring the doorbell "Tesla-gram" pause for response 🤖
Great Interview with James Douma, thanks a lot!!
ELON is a biased ass – the word is LIDAR ok
ppl dont write this SHT in C ! duh – maybe C++ or PYTHON/JAVA but NOT C ! where is this GUY FROM ?
the true "intervention rate" for FSD is not known because users take over in stretches of driving that FSDbeta is risky. So intervention rate could increase as FSD beta expands into more driving situations. Rate of decrease in interventions could also decrease due to the increasing difficulty of improvement as the problem areas become less frequent.
great video. Appreciate Dave and James doing this FSD/NN videos
Dave you are exceptionnal for doing what you have been doing with the Afghans. Hat's off to you.