Dave’s biggest value was in the guests he brought on (like this interview with James) He’s since been overtaken by Herbert who curates far better than Dave ever has and produces more quickly
James has no competition, is simply best at explaining complex AI world, in way we understand. Dave prompts well, that’s his role, it’s a special skill.
Yeah with Rob gone Dave is the last one that approaches the topic with the critical thinking and the distance needed. The rest has indeed largely formed into a cult.
We miss you Dave! With Rob now gone and Dillon being the only reliable frequent Tesla RUclipsr it would be great to have you back in the mix more often!!
At robotaxi day I’d love to see Robotaxi pull up to a charging station, have Optimus plug it in, fill the tire pressure, clean the headlamps, and change out a wiper blade. 😂
It is such a privilege that James shares his huge knowledge on so many subjects with us. Thank you for both of you for being examples of respectful, truthful people to look up to! 😇👏👏👏
I absolutely love when the right people meet up and are able to really peel back the layers of the Tesla / AI onion. I truly enjoyed the insights this particular duo were able to mine in a short 2 hour session. I am looking forward to rewatching this video to see how many more jewels I can tease out of the discourse that went over my head the first time around. Thanks to Dave and James for a wonderful immersion into the direction of current AI evolution and a peek at what may be just around the corner. 😊😊😊
Your one stop for FSD v12: Tesla's Autonomous Driving as it creates a whole new game. Thanks James Douma and Dave Lee. We know you are having fun with all this but your input is invaluable to us drivers and investors.
I needed a boring video to watch to put me to asleep to overcome jetlag but now I'll be up until 3am because I will be hanging on every word 😂😂 Thanks Dave & James - these are my favourite conversations to listen to, what a treat. I love knowing that as FSD gets superhuman you are the guys to tell us about just how good it's getting. Imagine how much content once Optimas is out there. I would love for Tesla to let you two be the first private individuals to get to experience Optimas (and Warren Redlich)
I agree with so many who have commented so far. It is wonderful to have the two of you back again! There is a rare something special that happens when you two put your hearts and minds together. Dave, your health seems better. I pray that is so and that your wife is getting well too. This is a fascinating conversation. I feel I am learning a bit about information theory, almost how to think! . xxox ramsay
Thanks for the interview with James. James provides such great descriptions that it makes End-to-end understandable. Please have him on whenever possible.
I'll add my thanks to you two for giving me the clearest view of what I'm experiencing using FSD 12.3.4. I was a skeptic about achieving autonomy via the heuristic code based path . I'm only beginning to understand the potential of machine learning with neural networks but man am I impressed with how my old (2018) Model X now can navigate the world. Crazy amazing times ...
Thanks guys, this is a very informative discussion for those of us (me, at least) who don't know anything much about machine learning but have a lot of experience with logic processes, say from software development in my case. Especially when a complex situation is illustrated by everyday situations we all encounter. The idea that humans work from a broad snapshot of a moment in time set in a certain situation like the blocked box is key. In an unconscious way we integrate all these inputs - green traffic light ahead, cross traffic has a red light, number of cars between me and the green light ahead, likelihood most will clear the lane or at least one will, etc. The result of the integration is a go-no-go decision. Once we go, we have changed the equation causing us to run a new integral that includes the input that our blocking the cross traffic allows for a possible delay in the forward traffic not impeding our goal of crossing the intersection. Whew! And all that in a few milliseconds.
Always an interesring interview and converssation with James. Thoroughly enjoyed this almost 2 hour video.I could listen to James discuss his views all day on the various topics he explains. Thanks Dave for having James on your channel
Please add sections to your videos and include them in your video description so we can click and skip to parts that interest us. It also gives us the ability determine if we should save this video for later or not. Thanks Dave.
Nova-Rift Dave is already working for free by not having ads. Now you want him to waste his precious time because you don’t want to watch the video? Rude
My experience with the free trial was surprising because I did not expect it to run straight into road closed sawhorses. You would think the cameras would at least notice the giant construction cranes and detour signs. Has worked great (other than potholes and too close to curbs) everywhere else for me.
I was surprised with some aspect. In some ways or was better than I expected but I have experienced some game stopper issues. - dirt roads, it is way too far toward the right to the point of being on the shoulder - three right hand veers on highway driving where I had to take over to prevent it trying right in front of cars (cutting them off) - didn't see a dirt pile the size of an overturned wheelbarrow through a right hand turn in a dirt road. For example, if there were a dog or a person laying in the ground it would have just plowed over them. Other than that it has been 99% no interventions.
❤ Awesome video. I remember when Dave said - It will take years before Tesla has solved FSD. Dave has always been positive. But that was before v12. So progress is super fast.
Great interview. Thanks for putting it up. Just to address your question about ‘guardrails’ in heuristic code v/s learning - there ARE guardrails in the training process as well - and they might get implemented as heuristic code - but it’s always part of any learning model .. and it’s called different things … but in LLMs it is called *policies* and lately also called *alignment* … Not saying I know more than James and a lot of my early understanding was based on some of his early explanations … but I think his response missed this key point
14:40 : for example going slightly off the side of the road a little when there's a drop off down a steep bank or cliff vs a field. Sometimes there isn't a barrier (Wales, UK , I'm looking at you)
Great conversation....helpful information. But I am very curious about the regressions and how to fix them if we've don't really know how they occur? How did Chuck's UPL suddenly fail hard with 3.4 update? Doesn't the AI build on itself? Also how can the brand new architecture make the exact same one-off mistakes that previous builds were making? How would 3.4 suddenly be acting like 11.2? Etc? Would be great to have you guys cover this.
Dave, thanks, one additional question I would love James and you to discuss is how Optimus supports Tesla’s mission? Every other product/decision has a clear line-of-site to the mission. Optimus is not as clear for me.
The iterative process: train a net, assess the result's nearness to the goal, change the weights, train again seems to be something like what an op amp integrator does when coupled with some sort of feedback. Is there a general term that describes the continuous integration of inputs over time?
I have seen 2 totals. Was so impressed with the seconds. But the first I was at the wrong age but in retrospect it was awesome. Totally in the middle of a National Wildlife refuge. 😢 second was as a real adult in northern Ga. Awesome so worth it.
FSD is GREAT ! Love it. Tesla should require viewing a short in car video to activate FSD. Tesla needs to open more delivery / service / collision centers. You should not have to drive hours to get your car serviced or repaired after an accident. Northern Indiana is a Tesla desert with the only repair / collision centers being in Chicago. What a traffic nightmare. Tesla needs to plan out service centers like they plan out charging stations. You should only have to drive so far to get to one. You should not have to travel to the next state to get your vehicle repaired. I know Tesla is trying to make vehicles need less service but accidents happen. Tesla needs more repair / collision centers so you do not have to drive to the next state to get your vehicle repaired. Having a large, delivery, service, collision center in Northern Indiana would spread out the overall service area. Many people are reluctant to buy a vehicle from a company that has no dealers / delivery /collision or repair centers within a reasonable driving distance.
Dave, v12 was a regression for me. I caution having this much enthusiasm, there is a lot of work before this is ready for prime time. One of main dangerous issues is the car, when deciding between lanes, tends to straddle both. This led my car to hurtle at high speed towards median trying to think whether to off ramp highway. Would have been a very bad accident. There are many not as dangerous interventions Ive had too.
Just drove 800 miles on fsd non stop other then stopping to charge and slept while car charged Let's just say I'm alive because of fsd. Very grateful to the Tesla engineers who figured this tech out. Cruise control is nice but having a car keep the car steady and change lanes and deal with construction areas. Definitely helps reduce the stress of driving long distances
@48:55 re: Parking lots... I think the major reason why parking lots aren't integrated yet is the complexity of logically mapping them in comparison to streets. We know Tesla is using a logical map to describe the world in a big picture way for route planning... Describing a parking lot in this way is so much more complicated.
FSD is making great progress. Tesla should allow owners to add FSD to a second owned Tesla family vehicle at a discounted price. Someone that has purchased FSD should have a transferable license and be able to use it on multiple family vehicles. Why has there been such a long delay to updating or refreshing the Model S and X interiors with the new ambient lighting. Many people are holding off buying a new Model S or X while waiting for the new features. Why the slow roll out ? The focus on FSD is great but the cars need timely updates to their features. The slow roll out hurts sales. New features need to be rolled out promptly and not dragged out. The lack of focus on the Model S and X is hurting volume sales. All the costs for research, development and amortization of manufacturing equipment for the S and X should be written off by now. It is time to lower the prices of the Model S and X to reflect the reduced costs.
I have been trying FSD V 12.3 for the last three weeks. Every time I got into a tricky situation, I just disabled and took over myself. It is going to be very challenging for Tesla to convince the regular people FSD is ready and capable, and to earn the trust. Just my 2 cents.
Yea I think that’s the common sense right now. I use it 97% of the time. 3% I just take over for a couple seconds if it looks like it may be tricky. Does surprisingly great in roundabouts by the way. Turns and stops well. I wish it would park and Summon to and from its parking spaces at a usable speed. Also would be nice if it changes lanes and slows to make room for emergency vehicles with sirens on. But needs sound sensors to deal with that and other car horns warning us of a hazard.
Hey fellas, Thank you both for such an amazing pod. Dave - w ould you be able to discuss inference computing prowess in a future pod? Just heard Chamath Palihapatiya interview the Founder of Groq, Jonathon Ross. He claims their "old school" approach to AI inference processing can break the monopoly held by Nvidia. He claimed that Nvidia has a strangle hold on the supply chain, which makes it impossible for any other alternative to compete with the architecture Nvidia uses. Would love you and James to discuss. Thanks for all you do...
… hi, Dave, I love your show watch it religiously whenever I see a new one. One thing, one recommendation I might have, is that stream of consciousness question scenario, which you employ, it’s just great, but it’s just too damn long :-)
My 2024 Model 3 LR has all steering button issues, they stop working randomly, 4 Tesla service visits and still a problem and getting worse. Tesla says it's a software bug, this is a serious safety issue. Help? 🙏
I don't see the "automatic doors" 🚪as a critical item. The customer will be charged 💸 until the car is fully operational for the next ride. ✅ We already have this today with bicycle or e-scooter renting and it works very well. 👍🏼
Sooo, does this new version (12.4) actually learn from the driver? I found the older versions to be too slow and indecisive (other drivers get annoyed). On one occasion it signal a left turn in the wrong left lane then cuts right into the center lane without a signal….almost an accident (would it be Teslas’ fault?). I just need FSD to drive around a dozen routes the way I’d drive them…yes, I’m willing to teach it!
No, and I doubt any versions will do that for the foreseeable future. FSD's training is centralized with Tesla and then sent to users as an update. Users do not have any input on training other than the data that was sent (which may or may not be used for the update)
I suspect Tesla will announce that private owners can join the robotaxi fleet. This will drive massive sales for the Model 3 and Model Y, along with FSD. Even once the robotaxi is built, it will still take several months to calibrate FSD for the new car and one to two years to build out the new robotaxi fleet.
Dave Lee and James Douma talk about "perception stack", "planning stack", "guardrails", "heuristics", "highway stack" and then mixed over and over with "end to end approach". Dave and James: this does not make sense: where are the "ends" in the "end to end"? Where are the modules with guardrails and heuristics? When are they activated? It would be wise, first to really extract the software structure (what modules, what stacks, what Neural networks, what call graph) from the Autopilot ECU in the different software versions (V12 and its subversions, V11 and its subversions) and hardware versions (Hw3 with front radar and ultrasonics, Hw3 with ultrasonics, Hw3 without ultrasonics, Hw4, Hw4 with front bumper camera), and then to visualize it, and then to talk about it. Otherwise, there is a lot of speculation about the internal structure, and the talk sounds like pseudo-tech intended to keep investors happy. The talk is still interesting but the amount of speculation is tremendous.
James is the Huberman of Machine Learning. I am so grateful that he takes this much time to educate us all, and on top of it all for free.
Definitely has more women tho
Guess you haven’t heard of the Huberman Scandal
Huberman can’t hold James’s jockstrap
@@maxpetee1690James’ partners at least know about each other lol
I heard James spun up some AI girlfriends and they are sandboxed.
James Douma is my favorite guest on your podcast! Love from 🇹🇼 to Dave and James! Thank you!
From where?!
@@capslock9031 Taiwan
The country, Taiwan 🇹🇼
Miss Dave’s TSLA input, glad to hear from James!
Dave’s biggest value was in the guests he brought on (like this interview with James)
He’s since been overtaken by Herbert who curates far better than Dave ever has and produces more quickly
@@frenchbully Dave’s deep insight and carefully weighed out thoughts are unparalleled.
@@Atgul-z6o if by “carefully weighed out thoughts” you mean taking 2-3weeks between videos than yeah I agree with you
I thought Dave was a bloke?
Glad to see your video again, Dave. Hope your family is well and healthy. Thanks for doing this
James has no competition, is simply best at explaining complex AI world, in way we understand.
Dave prompts well, that’s his role, it’s a special skill.
But the “competition is coming…” 😂
I miss these longform interviews from you Dave. Hope you and the family are doing well.
FINALLY Tesla youtubers I respect, that aren’t in a cult or greedy AF, and actually smart and competent ❤
People who throw around the word "cult" are actually the ones in a cult.
Yeah with Rob gone Dave is the last one that approaches the topic with the critical thinking and the distance needed. The rest has indeed largely formed into a cult.
@@lancejones4636Cult is when you disagree it seems currently…
@@lancejones4636 lmao, where do you get this from ? You seem bright, I’m sorry if I hurt your cult little feelings
@@Ample17 💯
Hey Dave, so good to see you!
Hope all is well with you and your family, thanks for doing this w/ James Douma, you guys are absolutely awesome.
Thank you for turning off ads. You’re the man Dave
Thank you gents!
Hey it’s one of the cult members mentioned in a previous comment!
My all time favorite RUclips duo.
We miss you Dave! With Rob now gone and Dillon being the only reliable frequent Tesla RUclipsr it would be great to have you back in the mix more often!!
I second that!!
Am glad to see Dave back and listen to James.
Am not sure that the perception stack is reliable yet with V12 that we got a few days ago🥺
At robotaxi day I’d love to see Robotaxi pull up to a charging station, have Optimus plug it in, fill the tire pressure, clean the headlamps, and change out a wiper blade. 😂
Optimus F1 pit crew
It is such a privilege that James shares his huge knowledge on so many subjects with us. Thank you for both of you for being examples of respectful, truthful people to look up to! 😇👏👏👏
Thanks Dave miss you guys together.
I absolutely love when the right people meet up and are able to really peel back the layers of the Tesla / AI onion. I truly enjoyed the insights this particular duo were able to mine in a short 2 hour session. I am looking forward to rewatching this video to see how many more jewels I can tease out of the discourse that went over my head the first time around. Thanks to Dave and James for a wonderful immersion into the direction of current AI evolution and a peek at what may be just around the corner. 😊😊😊
Your one stop for FSD v12: Tesla's Autonomous Driving as it creates a whole new game. Thanks James Douma and Dave Lee. We know you are having fun with all this but your input is invaluable to us drivers and investors.
I needed a boring video to watch to put me to asleep to overcome jetlag but now I'll be up until 3am because I will be hanging on every word 😂😂
Thanks Dave & James - these are my favourite conversations to listen to, what a treat.
I love knowing that as FSD gets superhuman you are the guys to tell us about just how good it's getting.
Imagine how much content once Optimas is out there. I would love for Tesla to let you two be the first private individuals to get to experience Optimas (and Warren Redlich)
Wahoo. James and Dave!🎉
Thanks for the great job. We sincerely appreciate your hard work and love hearing James. He is fantastic.
Great time for one of these. Thanks guys
Love the interview space..great set up. Another example of how independent content creators can beat the legacy media
Been looking forward to this video. Dave is a Smart guy and asks smart questions.
Damn, this dude knows his stuff and can communicate it clearly
Nice format Dave with this interview. Great questions, no nonsense as usual. Thanks to you and James for informing us.
I agree with so many who have commented so far. It is wonderful to have the two of you back again! There is a rare something special that happens when you two put your hearts and minds together. Dave, your health seems better. I pray that is so and that your wife is getting well too. This is a fascinating conversation. I feel I am learning a bit about information theory, almost how to think! . xxox ramsay
Thanks for the interview with James. James provides such great descriptions that it makes End-to-end understandable. Please have him on whenever possible.
So excited to see this episode.
I'll add my thanks to you two for giving me the clearest view of what I'm experiencing using FSD 12.3.4. I was a skeptic about achieving autonomy via the heuristic code based path . I'm only beginning to understand the potential of machine learning with neural networks but man am I impressed with how my old (2018) Model X now can navigate the world. Crazy amazing times ...
Love your interviews Dave!! Thanks for pulling in James as your chemistry is very natural and insights are valuable. Thank you guys!!
Thanks guys, this is a very informative discussion for those of us (me, at least) who don't know anything much about machine learning but have a lot of experience with logic processes, say from software development in my case. Especially when a complex situation is illustrated by everyday situations we all encounter. The idea that humans work from a broad snapshot of a moment in time set in a certain situation like the blocked box is key. In an unconscious way we integrate all these inputs - green traffic light ahead, cross traffic has a red light, number of cars between me and the green light ahead, likelihood most will clear the lane or at least one will, etc. The result of the integration is a go-no-go decision. Once we go, we have changed the equation causing us to run a new integral that includes the input that our blocking the cross traffic allows for a possible delay in the forward traffic not impeding our goal of crossing the intersection. Whew! And all that in a few milliseconds.
Thanks for another great conversation about this important subject. James is at another level. 👍
Always an interesring interview and converssation with James. Thoroughly enjoyed this almost 2 hour video.I could listen to James discuss his views all day on the various topics he explains. Thanks Dave for having James on your channel
I like this podcast format
So happy to see this video. I hope it delivers!
Love these chats! ❤
Christmas is here! Finally Dave and James back on the screen discussing FSD ❤❤
Great seeing you guys together again!
Please add sections to your videos and include them in your video description so we can click and skip to parts that interest us. It also gives us the ability determine if we should save this video for later or not. Thanks Dave.
Nova-Rift Dave is already working for free by not having ads. Now you want him to waste his precious time because you don’t want to watch the video? Rude
My favorite two people, thank you for another great discussion! Listened to it while coming back from a road trip on FSD
I want to see both of you more often!
Thank you Dave and James. This is an outstanding discussion.
My experience with the free trial was surprising because I did not expect it to run straight into road closed sawhorses. You would think the cameras would at least notice the giant construction cranes and detour signs. Has worked great (other than potholes and too close to curbs) everywhere else for me.
I was surprised with some aspect. In some ways or was better than I expected but I have experienced some game stopper issues.
- dirt roads, it is way too far toward the right to the point of being on the shoulder
- three right hand veers on highway driving where I had to take over to prevent it trying right in front of cars (cutting them off)
- didn't see a dirt pile the size of an overturned wheelbarrow through a right hand turn in a dirt road. For example, if there were a dog or a person laying in the ground it would have just plowed over them.
Other than that it has been 99% no interventions.
❤ Awesome video. I remember when Dave said - It will take years before Tesla has solved FSD. Dave has always been positive. But that was before v12. So progress is super fast.
James is one of my favorite to listen to
Thank you! What a wonderful discussion.
brilliant and sooo instructive - thanks a lot!
Always in for a fantastic video when James is around.
Really good conversation 👏🏼
Loving that plant on the table.
The plant on the table is the most important thing in the whole set. Obviously.
Love your videos with James!
This studio is bad ass
Best content on the internet
Nice conversation.
Love James and you doing video, you talk so the everyday person can understand
Substantially increased my understanding of FSD mechanics!
Oh no! I have to go to bed, and then this just dropped....
The goat is back!
Great interview. Thanks for putting it up.
Just to address your question about ‘guardrails’ in heuristic code v/s learning - there ARE guardrails in the training process as well - and they might get implemented as heuristic code - but it’s always part of any learning model .. and it’s called different things … but in LLMs it is called *policies* and lately also called *alignment* …
Not saying I know more than James and a lot of my early understanding was based on some of his early explanations … but I think his response missed this key point
Would love timestamps. Good to see you again Dave!
14:40 : for example going slightly off the side of the road a little when there's a drop off down a steep bank or cliff vs a field. Sometimes there isn't a barrier (Wales, UK , I'm looking at you)
Excellent discussion. Very informative and thought provoking. Elon's presentation on 8/8 will be very interesting, which I will watch intently!
Great discussion on a super interesting topic. Thanks!
Worth listen to twice!
On top of it all, great production! Three cameras makes a big difference.
Downloading 2024.3.15 as I listen to this. Super excited!!
Wahoo great to see you guys again - the Originals !
very nice and relaxed setup 🙂
Missed you guys. I can not believe there are only 35k views in 24 hours, the population of one small town for something that WILL change the world.
Great conversation....helpful information. But I am very curious about the regressions and how to fix them if we've don't really know how they occur? How did Chuck's UPL suddenly fail hard with 3.4 update? Doesn't the AI build on itself?
Also how can the brand new architecture make the exact same one-off mistakes that previous builds were making? How would 3.4 suddenly be acting like 11.2? Etc? Would be great to have you guys cover this.
Dave, thanks, one additional question I would love James and you to discuss is how Optimus supports Tesla’s mission? Every other product/decision has a clear line-of-site to the mission. Optimus is not as clear for me.
The iterative process: train a net, assess the result's nearness to the goal, change the weights, train again seems to be something like what an op amp integrator does when coupled with some sort of feedback. Is there a general term that describes the continuous integration of inputs over time?
Love your T-Shirt James 🤓
Thanks Dave ! As usual an impressive FSD content and in great detail.
I think the supervised robotaxi trial is a must as simply an education push to the broader public.
Isn't Amazon a bookstore?
I have seen 2 totals. Was so impressed with the seconds. But the first I was at the wrong age but in retrospect it was awesome. Totally in the middle of a National Wildlife refuge. 😢 second was as a real
adult in northern Ga. Awesome so worth it.
The dreamteam again. TY
FSD is GREAT ! Love it. Tesla should require viewing a short in car video to activate FSD.
Tesla needs to open more delivery / service / collision centers. You should not have to drive hours to get your car serviced or repaired after an accident. Northern Indiana is a Tesla desert with the only repair / collision centers being in Chicago. What a traffic nightmare. Tesla needs to plan out service centers like they plan out charging stations. You should only have to drive so far to get to one. You should not have to travel to the next state to get your vehicle repaired.
I know Tesla is trying to make vehicles need less service but accidents happen. Tesla needs more repair / collision centers so you do not have to drive to the next state to get your vehicle repaired. Having a large, delivery, service, collision center in Northern Indiana would spread out the overall service area.
Many people are reluctant to buy a vehicle from a company that has no dealers / delivery /collision or repair centers within a reasonable driving distance.
Thanks for this!
Dave, v12 was a regression for me. I caution having this much enthusiasm, there is a lot of work before this is ready for prime time. One of main dangerous issues is the car, when deciding between lanes, tends to straddle both. This led my car to hurtle at high speed towards median trying to think whether to off ramp highway. Would have been a very bad accident. There are many not as dangerous interventions Ive had too.
Highway driving is still using v11 heuristics.
@@DC-yh2oyAgreed. Alpha Office needs to do a little more research to actually understand FSD.
The spider plant is a nice touch.
Just drove 800 miles on fsd non stop other then stopping to charge and slept while car charged
Let's just say I'm alive because of fsd. Very grateful to the Tesla engineers who figured this tech out. Cruise control is nice but having a car keep the car steady and change lanes and deal with construction areas. Definitely helps reduce the stress of driving long distances
@48:55 re: Parking lots... I think the major reason why parking lots aren't integrated yet is the complexity of logically mapping them in comparison to streets. We know Tesla is using a logical map to describe the world in a big picture way for route planning... Describing a parking lot in this way is so much more complicated.
FSD is making great progress. Tesla should allow owners to add FSD to a second owned Tesla family vehicle at a discounted price.
Someone that has purchased FSD should have a transferable license and be able to use it on multiple family vehicles.
Why has there been such a long delay to updating or refreshing the Model S and X interiors with the new ambient lighting.
Many people are holding off buying a new Model S or X while waiting for the new features. Why the slow roll out ?
The focus on FSD is great but the cars need timely updates to their features. The slow roll out hurts sales.
New features need to be rolled out promptly and not dragged out. The lack of focus on the Model S and X is hurting volume sales.
All the costs for research, development and amortization of manufacturing equipment for the S and X should be written off by now.
It is time to lower the prices of the Model S and X to reflect the reduced costs.
When does fsd start running the semi? Will that be a different stack?
Great vid.
I have been trying FSD V 12.3 for the last three weeks. Every time I got into a tricky situation, I just disabled and took over myself. It is going to be very challenging for Tesla to convince the regular people FSD is ready and capable, and to earn the trust. Just my 2 cents.
that is a smart strategy.
Yea I think that’s the common sense right now. I use it 97% of the time. 3% I just take over for a couple seconds if it looks like it may be tricky. Does surprisingly great in roundabouts by the way. Turns and stops well. I wish it would park and Summon to and from its parking spaces at a usable speed. Also would be nice if it changes lanes and slows to make room for emergency vehicles with sirens on. But needs sound sensors to deal with that and other car horns warning us of a hazard.
Hey fellas, Thank you both for such an amazing pod. Dave - w ould you be able to discuss inference computing prowess in a future pod? Just heard Chamath Palihapatiya interview the Founder of Groq, Jonathon Ross. He claims their "old school" approach to AI inference processing can break the monopoly held by Nvidia. He claimed that Nvidia has a strangle hold on the supply chain, which makes it impossible for any other alternative to compete with the architecture Nvidia uses. Would love you and James to discuss. Thanks for all you do...
Quality content.
… hi, Dave, I love your show watch it religiously whenever I see a new one. One thing, one recommendation I might have, is that stream of consciousness question scenario, which you employ, it’s just great, but it’s just too damn long :-)
My 2024 Model 3 LR has all steering button issues, they stop working randomly, 4 Tesla service visits and still a problem and getting worse. Tesla says it's a software bug, this is a serious safety issue. Help? 🙏
I don't see the "automatic doors" 🚪as a critical item. The customer will be charged 💸 until the car is fully operational for the next ride. ✅
We already have this today with bicycle or e-scooter renting and it works very well. 👍🏼
Sooo, does this new version (12.4) actually learn from the driver? I found the older versions to be too slow and indecisive (other drivers get annoyed). On one occasion it signal a left turn in the wrong left lane then cuts right into the center lane without a signal….almost an accident (would it be Teslas’ fault?). I just need FSD to drive around a dozen routes the way I’d drive them…yes, I’m willing to teach it!
No, and I doubt any versions will do that for the foreseeable future. FSD's training is centralized with Tesla and then sent to users as an update. Users do not have any input on training other than the data that was sent (which may or may not be used for the update)
Are the guard rails off? Who said that?
I suspect Tesla will announce that private owners can join the robotaxi fleet. This will drive massive sales for the Model 3 and Model Y, along with FSD.
Even once the robotaxi is built, it will still take several months to calibrate FSD for the new car and one to two years to build out the new robotaxi fleet.
What is the deal with the Cybertruck FSD?
Came here for FSD, stayed for Gen AI news.
Came for FSD, Left for work
Dave Lee and James Douma talk about "perception stack", "planning stack", "guardrails", "heuristics", "highway stack" and then mixed over and over with "end to end approach". Dave and James: this does not make sense: where are the "ends" in the "end to end"? Where are the modules with guardrails and heuristics? When are they activated? It would be wise, first to really extract the software structure (what modules, what stacks, what Neural networks, what call graph) from the Autopilot ECU in the different software versions (V12 and its subversions, V11 and its subversions) and hardware versions (Hw3 with front radar and ultrasonics, Hw3 with ultrasonics, Hw3 without ultrasonics, Hw4, Hw4 with front bumper camera), and then to visualize it, and then to talk about it. Otherwise, there is a lot of speculation about the internal structure, and the talk sounds like pseudo-tech intended to keep investors happy. The talk is still interesting but the amount of speculation is tremendous.
No timestamps? 😢