I really enjoyed this conversation with Sertac. Here's the outline: 0:00 - Introduction 1:44 - Autonomous flying vs autonomous driving 6:37 - Flying cars 10:27 - Role of simulation in robotics 17:35 - Game theory and robotics 24:30 - Autonomous vehicle company strategies 29:46 - Optimus Ride 47:08 - Waymo, Tesla, Optimus Ride timelines 53:22 - Achieving the impossible 53:50 - Iterative learning 58:39 - Is Lidar is a crutch? 1:03:21 - Fast autonomous flight 1:18:06 - Most beautiful idea in robotics
So, even as a computer science graduate I'm not nearly as experienced as an MIT professor, but you might want to remind him that computer science has struggled with the problem of predicting the future before in the form of CPU schedulers. The answer ended up being that it is impossible to know the future for sure, and you have to settle for only accounting for the most likely possibilities and reacting as issues appear.
Lex, I am big fan of your podcast. Very honest, straight forward questions that brings out the best knowledge of the guest. I cannot thank you enough for educating the audiences.
@@zeppelincheetah It's basically a popular saying in Turkish social media. It means "hang the flags". It is used in a tongue in cheek way, when some Turkish person does something nice, makes an appearance on the international arena, etc.
Rating: 6.9/10 In Short: Futuristic Flying (and driving) Focused Notes: This convo was a lot more focused than many of the future lex pods with AI related people, which I think givin the 1.5 hour length was actually good. Neat to think about a future where we have self (or semi autonomous) flying vehicles that take us from boston to NYC, a thought experiment that they talk about a lot here. Sertac is a very humble, honest, and clear speaker, but a bit of a nerd (in a good way). The whol eidea of optimus ride was talked about in the perfect amount of detail and super cool, but interesting how this was 4 years ago from the time I've watched it and I still havnt heard much from this. Seems like the timing of a lot of these kinds of ideas/companys is less on the scale of years and more like decades. Would be cool to have him on again now and get more of his story and background and purpose.
*_1_* With his rich baritone voice, excellent American English pronunciation, and deep questions, Lex Fridman is the best interviewer on TH-cam. *_2_* 1:11...Right on target, Lex! 👏 Are you into cryptos? *_3_* Sertac Karaman is brilliant, as all MIT professors are. 👍 💕 ☮ 🌎 🌌
Sentac is obviously a smart person. When you asked about timelines, his answer was anchored around the concept of technology gaps. He says those are fundamentally unpredictable. He says Musk predicts them. Then he concludes, basically, Musk is predicting the unpredictable. Ray Kurzweil accurately predicted the solving time for the human genome. I think the reason for this conflict of understanding is that Sentac does not have an intuitive sense of exponential change. In fact, humans are basically programmed to intuitively predict in a lineal way not an exponential way. Musk and Kurzwell make their predictions because, unlike the vast majority of people, they do have an intuitive exponential sense this way. But then of course Musk is usually too optimistic. This is because the “sense of exponential change” is not as accurate as the lineal sense people are accustomed to.
I think Taskin Padir from northeastern or Greg Fischer from WPI would make for other great robotics discussions. I've had the pleasure to talk and work with them and Sertac. All great researchers.
Talking about air and road many people are missing the opportunity in ships and boats. They already come equipped with expensive lidar and radar. Also a captain costs more than driver or a pilot.
Autonomy versus sustainability... Once you have effective autonomy and prove it safer than human driving, it can only lead to banning humans from driving (with some allowed exceptions). But even before and at the early stages of introduction, autonomy will by definition be sustainable because it will save human time a very valuable commodity.
so what should we expect ? to have thousands of objects flying in the sky above us ? The blue sky obscured by all that opacity ? How could we watch the sky without having an IFO in the screen ?
Oh how very clever these people are! Utilising every bit of space possible is advancement!? I beg to differ. We are in an age where the cerebral cortex is dominating, while the heart sits trapped in a cage 😓
Look into convolutional neural networks. They are the systems on which “perception” equations or matrices operate. Basically, the camera captures an image, the image is scanned at full resolution, half, quarter, etc. until your image is just a single pixel. At one or many layers of abstraction, the data created from the original image is compared with the layers of data from a large set of layers data of relevant images to predict the contents of the image, the distance of the objects in the image, etc. This is combined with data from radar and laser scanning of the surrounding environment, and the data is compounded and helps produce much more accurate distance and object size measuring. I hope this answers your question.
I bet your guy is very smart but please send him back to physics 101. Lifting an object vs rolling an object is a 10 fold increment in energy consumption. Our society is starving for energy. Imagining a future with individual flying cars is just crazy. Only the richest Manhattan citizens would dream about it.
Compute power coupled with precise reckoning will dominate the application of safety in the air. Think about hyper precise location (you know within an inch where all objects are with appropriate margins for dynamic objects...e.g. tree limbs in the wind). Now, couple that with real time tracking/simulation. Automatic flight control can take advantage provided corridors (volume lanes in the air). Anyway, rules and constraints coupled with precise location and margins of error, is not that compute intensive. If every vehicle in essence is coupled and has its own synced controlled flight, safety should fall out. We could probably do it with today's technology, but it will get easier and easier as compute power increases. AI's can invent the necessary rules and ensure safety margins.
i love all the topics discussed here, but the guest is not really knowledgable about most of those high level questions. for example he doesn't understand economics/business but tries to explain and predict things that can only be explained from an economics/business perspective. I'm sure he has some extremely good technical knowhow, but unfortunately the conversation never got very technical
I agree with you, of course lex is doing a great job, but people are exaggerating maybe because of ignorance and not even trying to think critically for improving.
Unfortunately I don't think David does much with CS/AI, and although his work is highly relevant to CS/AI, he has not explored that. The work he is doing in genetics do have a lot to do with very general reward functions and are related to long term memory, but he hasn't really explored that. I'd worry that Lex and David would spend an hour on the verge of having a really interesting conversation, but couldn't quite bridge the gap between their respective fields any more than David already has on JRE or in his own podcast.
Get callen on the fucking show. you have genius after genius....you could have christ on the podcast or einstein but how many of them can make you laugh.
I really enjoyed this conversation with Sertac. Here's the outline:
0:00 - Introduction
1:44 - Autonomous flying vs autonomous driving
6:37 - Flying cars
10:27 - Role of simulation in robotics
17:35 - Game theory and robotics
24:30 - Autonomous vehicle company strategies
29:46 - Optimus Ride
47:08 - Waymo, Tesla, Optimus Ride timelines
53:22 - Achieving the impossible
53:50 - Iterative learning
58:39 - Is Lidar is a crutch?
1:03:21 - Fast autonomous flight
1:18:06 - Most beautiful idea in robotics
So, even as a computer science graduate I'm not nearly as experienced as an MIT professor, but you might want to remind him that computer science has struggled with the problem of predicting the future before in the form of CPU schedulers. The answer ended up being that it is impossible to know the future for sure, and you have to settle for only accounting for the most likely possibilities and reacting as issues appear.
Glad to hear you enjoyed it, Lex. You always enjoy it.
I love that youtube now integrates the outline in the video itself.
What are best books to read for programming?
One of the best intellectual podcasts around. Thanks so much Lex!
Greetings from Turkey 👌 Thank you
Great discussion. Thank you again, Lex. Could we please have Andrej Karpathy on this show?
THIS!!!!
Book it Lex!
This!
Please bring Karpathy !
We are definitely proud of him and see him as a mentor too! Love from 🇹🇷💜
🇹🇷🇹🇷🇹🇷
Lex, I am big fan of your podcast. Very honest, straight forward questions that brings out the best knowledge of the guest. I cannot thank you enough for educating the audiences.
İnsan bir ayrı mutlu oluyor
Türk var mı türk ?
Olmaz mi :)
var hocam
thanks Lex .
Sertac Karaman adamsin .
1:21:16 literally gave me goosebumps!
He didn't mention the issue of noise when it comes to making flying cars. We don't only need to make it safe, we also need to make it very silent.
As bayrakları as
What does that mean? Google translate says "ace flags"
@@zeppelincheetah It's basically a popular saying in Turkish social media. It means "hang the flags". It is used in a tongue in cheek way, when some Turkish person does something nice, makes an appearance on the international arena, etc.
That was great. The fact that we are at the physical limits on many of these technologies was really interesting thing to think about.
Lex looks happier every video, Looking good Lex!
Rating: 6.9/10
In Short: Futuristic Flying (and driving) Focused
Notes: This convo was a lot more focused than many of the future lex pods with AI related people, which I think givin the 1.5 hour length was actually good. Neat to think about a future where we have self (or semi autonomous) flying vehicles that take us from boston to NYC, a thought experiment that they talk about a lot here. Sertac is a very humble, honest, and clear speaker, but a bit of a nerd (in a good way). The whol eidea of optimus ride was talked about in the perfect amount of detail and super cool, but interesting how this was 4 years ago from the time I've watched it and I still havnt heard much from this. Seems like the timing of a lot of these kinds of ideas/companys is less on the scale of years and more like decades. Would be cool to have him on again now and get more of his story and background and purpose.
Teşekkürler Lex.
As always, I didn't realize time passing when listening to your podcast. The best AI podcast. Thank you Lex.
*_1_* With his rich baritone voice, excellent American English pronunciation, and deep questions, Lex Fridman is the best interviewer on TH-cam.
*_2_* 1:11...Right on target, Lex! 👏 Are you into cryptos?
*_3_* Sertac Karaman is brilliant, as all MIT professors are. 👍
💕 ☮ 🌎 🌌
Great questions as always, thank you for the episode 👌
Thanks Lex! These interviews is key to have Insights.
As bayrakları
Asin bayraklari
Thank you for doing this Lex. It is just amazing. What do you think about Naval Ravikant on your podcast? it would be mind-blowing
As bayraklari
Nice, thanks!
Wow 😲 “brilliant” thank you for this interview I like how he respect competitors
Love it when Lex's guests aren't too cool to have fun.
When are you moving to Spotify?
Last time I was this early, Arthur Samuel coined the phrase "Machine Learning"
it'ssss awesome!!!! Great time here!
Sentac is obviously a smart person. When you asked about timelines, his answer was anchored around the concept of technology gaps. He says those are fundamentally unpredictable.
He says Musk predicts them. Then he concludes, basically, Musk is predicting the unpredictable.
Ray Kurzweil accurately predicted the solving time for the human genome.
I think the reason for this conflict of understanding is that Sentac does not have an intuitive sense of exponential change. In fact, humans are basically programmed to intuitively predict in a lineal way not an exponential way. Musk and Kurzwell make their predictions because, unlike the vast majority of people, they do have an intuitive exponential sense this way.
But then of course Musk is usually too optimistic. This is because the “sense of exponential change” is not as accurate as the lineal sense people are accustomed to.
Towards the end, one of his real-seeming timeline guesses for mature systems that can use cameras only was *fifty years*. like come on dude.
Lex, you need to step up and fill the void Joe's gonna leave in YT, great podcasts
I think Taskin Padir from northeastern or Greg Fischer from WPI would make for other great robotics discussions. I've had the pleasure to talk and work with them and Sertac. All great researchers.
You rock my world!!!! Love you man
It would be great if you got Noonian Soong on your podcast.
Can anyone point me to resources concerning these amazing advances in camera simulation that was mentioned near the start?
Talking about air and road many people are missing the opportunity in ships and boats. They already come equipped with expensive lidar and radar. Also a captain costs more than driver or a pilot.
Saygilar Sertac Hocam. Almanydan Selamlar.
Sertac is too close to Serac, anyone who has seen Westworld should be worried lol
hhahaha
Westworld sertac
Autonomy versus sustainability... Once you have effective autonomy and prove it safer than human driving, it can only lead to banning humans from driving (with some allowed exceptions). But even before and at the early stages of introduction, autonomy will by definition be sustainable because it will save human time a very valuable commodity.
Do Charles Elkan!
Optimus Drive > Optimus Ride
Mechanical things break. When they break, they fall out of the sky. Raining metal.
Sertac Hard Carry
so what should we expect ? to have thousands of objects flying in the sky above us ? The blue sky obscured by all that opacity ? How could we watch the sky without having an IFO in the screen ?
Oh how very clever these people are! Utilising every bit of space possible is advancement!? I beg to differ. We are in an age where the cerebral cortex is dominating, while the heart sits trapped in a cage 😓
Setrac is working on the right set of problems for success in his field
Should address the UFO videos lex
👏🏻👏🏻👏🏻👏🏻👏🏻👏🏻👏🏻👏🏻
neat episode buddy . dig your show . cheers yo #funwithbrainlesions #beercuresMS
...And Robots that climb on rocks.
Great question let me tell you this and i will get to that....
14:31 something is definitely off in THIS video
Does someone have some Reading Material, or source recommendations for "perception in computer science." ? Thank you
Look into convolutional neural networks. They are the systems on which “perception” equations or matrices operate. Basically, the camera captures an image, the image is scanned at full resolution, half, quarter, etc. until your image is just a single pixel. At one or many layers of abstraction, the data created from the original image is compared with the layers of data from a large set of layers data of relevant images to predict the contents of the image, the distance of the objects in the image, etc. This is combined with data from radar and laser scanning of the surrounding environment, and the data is compounded and helps produce much more accurate distance and object size measuring. I hope this answers your question.
👏
Love this one. Especially after watching Westworld
I have cash app. I love it.
Is he Turkish or Persian?
Turkish
Let's agree on human here. I think those other two labels are there to describe nations, much less to describe a human being.
I bet your guy is very smart but please send him back to physics 101.
Lifting an object vs rolling an object is a 10 fold increment in energy consumption. Our society is starving for energy. Imagining a future with individual flying cars is just crazy. Only the richest Manhattan citizens would dream about it.
Compute power coupled with precise reckoning will dominate the application of safety in the air. Think about hyper precise location (you know within an inch where all objects are with appropriate margins for dynamic objects...e.g. tree limbs in the wind). Now, couple that with real time tracking/simulation. Automatic flight control can take advantage provided corridors (volume lanes in the air). Anyway, rules and constraints coupled with precise location and margins of error, is not that compute intensive. If every vehicle in essence is coupled and has its own synced controlled flight, safety should fall out. We could probably do it with today's technology, but it will get easier and easier as compute power increases. AI's can invent the necessary rules and ensure safety margins.
i love all the topics discussed here, but the guest is not really knowledgable about most of those high level questions. for example he doesn't understand economics/business but tries to explain and predict things that can only be explained from an economics/business perspective. I'm sure he has some extremely good technical knowhow, but unfortunately the conversation never got very technical
I agree with you, of course lex is doing a great job, but people are exaggerating maybe because of ignorance and not even trying to think critically for improving.
Why not fake that autonomous vehicles have a human driver, perhaps with a hologram.
Lex pls bring Stanford's Robert Sapolsky
we are taking a lot of tangents today
welcome to the AI Podcast
woa 6 seconds ago
You should invite David Sinclair, He is amazing. thank you lex
Unfortunately I don't think David does much with CS/AI, and although his work is highly relevant to CS/AI, he has not explored that.
The work he is doing in genetics do have a lot to do with very general reward functions and are related to long term memory, but he hasn't really explored that. I'd worry that Lex and David would spend an hour on the verge of having a really interesting conversation, but couldn't quite bridge the gap between their respective fields any more than David already has on JRE or in his own podcast.
Why we see fly car ? I think that we see in next 50 years.
Get callen on the fucking show. you have genius after genius....you could have christ on the podcast or einstein but how many of them can make you laugh.
Flying cars is a STUPID idea. Period.
Pretty boring conversation.
Turkish pushy
One of the best intellectual podcasts around. Thanks so much Lex!