The Future is UNICORNS and RAINBOWS! James Douma on Tesla FSD Beta, LLMs, the AI Singularity & more!
ฝัง
- เผยแพร่เมื่อ 12 พ.ค. 2023
- James Douma returns to the channel in a big way, discussing just how far Tesla's FSD Beta has come, the wild evolution of Large Language Models (LLMs like ChatGPT and Llama) and Diffusion models (like Midjourney and Dall-E 2) in the AI space, and the challenges we all face--from our jobs to our very existence--with Artificial General Intelligence (AGI) potentially soon!
Follow James on Twitter here: / jamesdouma
Join this channel to get access to perks:
/ @drknowitallknows
**To become part of our Patreon team, help support the channel, and get awesome perks, check out our Patreon site here: / drknowitallknows . Thanks for your support!
Go to athleticgreens.com/drknowitall to get started on your first purchase and receive a FREE 1-year supply of Vitamin D3+K2 and 5 travel packs
Get 25% off Blinkist premium and enjoy 2 memberships for the price of 1! Start your 7-day free trial by clicking here: blinkist.com/drknowitallknows
Get The Elon Musk Mission (I've got two chapters in it) here:
Paperback: amzn.to/3TQXV9g
Kindle: amzn.to/3U7f7Hr!
**Want some awesome Dr. Know-it-all merch, including the AI STUDENT DRIVER Bumper Sticker? Check out our awesome Merch store: drknowitall.itemorder.com/sale
For a limited time, use the code "Knows2021" to get 20% off your entire order!
**Check out Artimatic: www.artimatic.io
**For 30% off the amazing OneCarStereo instrument display for Tesla Model 3/Y, use this link: bit.ly/3vd7NRv
**Want to get in on WeBull's awesome Crypto and stock fun? Check out this link, and get started trading stock and Crypto!
a.webull.com/i/DrKnow-it-allK...
**If you are looking to purchase a new Tesla CAR, Solar roof, Solar tiles or PowerWall, just click this link to get up to $500 off! www.tesla.com/referral/john11286. Thank you!
**You can help support this channel with one click! We have an Amazon Affiliate link in several countries. If you click the link for your country, anything you buy from Amazon in the next several hours gives us a small commission, and costs you nothing. Thank you!
* USA: amzn.to/39n5mPH
* Germany: amzn.to/2XbdxJi
* United Kingdom: amzn.to/3hGlzTR
* France: amzn.to/2KRAwXh
* Spain: amzn.to/3hJYYFV
**What do we use to shoot our videos?
-Sony alpha a7 III: amzn.to/3czV2XJ
--and lens: amzn.to/3aujOqE
-Feelworld portable field monitor: amzn.to/38yf2ah
-Neewer compact desk tripod: amzn.to/3l8yrUk
-Glidegear teleprompter: amzn.to/3rJeFkP
-Neewer dimmable LED lights: amzn.to/3qAg3oF
-Rode Wireless Go II Lavalier microphones: amzn.to/3eC9jUZ
-Rode NT USB+ Studio Microphone: amzn.to/3U65Q3w
-Focusrite Scarlette 2i2 audio interface: amzn.to/3l8vqDu
-Studio soundproofing tiles: amzn.to/3rFUtQU
-Sony MDR-7506 Professional Headphones: amzn.to/2OoDdBd
-Apple M1 Max Studio: amzn.to/3GfxPYY
-Apple M1 MacBook Pro: amzn.to/3wPYV1D
-Docking Station for MacBook: amzn.to/3yIhc1S
-Philips Brilliance 4K Docking Monitor: amzn.to/3xwSKAb
-Sabrent 8TB SSD drive: amzn.to/3rhSxQM
-DJI Mavic Mini Drone: amzn.to/2OnHCEw
-GoPro Hero 9 Black action camera: amzn.to/3vgVMrH
-GoPro Max 360 camera: amzn.to/3nORGYk
-Tesla phone mount: amzn.to/3U92fl9
-Suction car mount for camera: amzn.to/3tcUfRK
-Extender Rod for car mount camera: amzn.to/3wHQXsw
**Here are a few products we've found really fun and/or useful:
-NeoCharge Dryer/EV charger splitter: amzn.to/39UcKWx
-Lift pucks for your Tesla: amzn.to/3vJF3iB
-Emergency tire fill and repair kit: amzn.to/3vMkL8d
-CO2 Monitor: amzn.to/3PsQRh2
-Camping mattress for your Tesla model S/3/X/Y: amzn.to/3m7ffef
**Music by Zenlee. Check out his amazing music on instagram -@zenlee_music
or TH-cam - / @zenlee_music
Tesla Stock: TSLA
**EVANNEX
Check out the Evannex web site: evannex.com/
If you use my discount code, KnowsEVs, you get $10 off any order over $100!
**For business inquiries, please email me here: DrKnowItAllKnows@gmail.com
Twitter: / drknowitall16
Also on Twitter: @Tesla_UnPR: / tesla_un
Instagram: @drknowitallknows
**Want some outdoorsy videos? Check out Whole Nuts and Donuts: / @wholenutsanddonuts5741 - วิทยาศาสตร์และเทคโนโลยี
chapters please🙏
1:16 When will FSD be ready?
7:10 Tesla claiming liability & FSD confidence
9:16 The path forward: Challenges Developing FSD
10:14 Waymo & Cruise
15:19 v11.4 Release notes, Planning
20:53 10.69 Behavior in Austin
21:57 v11 interventions
26:00 v12
30:00 Human behavior in driving.
31:58 11.4 controls stack
33:30 v12- End to End AI
34:14 The craziness of LLMs
49:39 What is AI’s next big development?
59:34 The Future of Employment for Humanity
1:07:07 The Singularity. Is AI going to take over?
We need AI to do this automatically
Always a joy to be able to listen to James!
he has a very large stack. he goes down all these rabbit holes, backs out and always ends up where he wanted.
Love James Douma.
Dr Knows it All is a poser and self-indulged personality with desires for front row tickets, in a crowd of more knowledgeable people.
James seems a patient and loving man in a whiny conversation.
😊
I will Always tune in when James is on. One of the main reasons I became a Tesla Bull.....
"Learn to keep learning" ~ Dr. Know-It-All 2023
James Douma seems a very knowledgable guest. Thanks both.
I totally love interviews with James Douma. The future sure sounds a lot brighter now. Best wishes for your children doc. Thanks so much!! 🙏 Lotsa love to all as always ❤❤❤
Seriously, we are so lucky to have James!! Great conversation!!!
My vote: this was one of your best interviews. I love exploring the possibilities. Hearing your discussion with James on these topics is brain candy!!
Love the story at the end about a ChatGPT negotiating for a lower cable price with a cable company agent to get a lower cable bill and the cable company chat agent is also is a ChatGPT!
I think this was the best Know it All episode of all time.
AI is a deep rabbit hole to go down into. It's great to have you and James as guides. The world is changing and doing so at a rate we can't really keep up with. To those of us who understand that it's both exciting and frightening all at the same time. Thanks for all the time you spend trying to help us understand. Tell James thanks as well his wide knowledge of the big picture is unique. He's a generalist with very specific knowledge in key areas. That's almost unheard of and a treasure to find and have access to. This journey we are on is certainly unlike anything we could have even dreamed of just a few years ago.
I can listen to the two of you discussing for hours!
That was a really awesome chat. Well done both of you!
I am glad James shares his insights in these topics. Interesting times ahead
I normally shy away from videos this long. James Douma made this video extremely interesting.
Many things to think about from this conversation. Thank you.
TY for sharing, and tx to James for believing in a better world.... this is where I start to doubt. But we will see. Love and hugs to the both of you from Vienne/Autriche .
James when you spoke about Chinese and US AI systems competing in the future I had a hard flashback to "Neuromancer" a book by William Gibson about AI and computer viruses fighting in the future. Sci-Fi literature 30+ years ago predicting the future in uncanny ways.
Excellent discussion on AI.. Enjoyed very much.
I installed 11.4.1 last night. Eager to test it out. I was surprised to have it so soon.
Thank you for the Skyscraper analogy. Very helpful.
Imagine thirty years ago trying to predict the societal effect of the internet and cell phone.
It would have been a fool’s errand.
So it is with predicting the future of AGI.
We want more of that. Awesome.
Great conversation!
@1:38 excellent statement James. Thanks
Sharp hills, sharp curves with disappearing or obscured road, is exactly the problem autosteer and FSD continues to chew on here in northern New England in my 2023 MY. The first time autosteer sees a new road and does the catastrophically wrong thing (IE, choosing a route straight off the road at extremely dangerous location) is a *not fun* moment - takes every bit of personal willpower to not intervene until the last last last moment. But I keep running the Y through the exercise and am amazed to see the car improving in every situation.
And a big thanks to JOHN, for getting better and better keeping his yapper closed except to add valuable content. Much more watchable video.
How about changing roads every few meters?
Wide, narrow, a house in the way, a though house aka small tunnel, totally obscured walkways which you have to cross, single lane, 1/2 lanes, 2 lanes and single lane again without markings? We got many of those here in Austria.
Or how about the road passing a walkway where the cars have to slow to 5 km/h without any sign?
The car is not learning by repetition.
All the compute is done at Tesla from information relayed by your intervention or its background monitoring in ghost mode. Improvement comes back as a new version download.
@@waynerussell6401 Understood and agree. I could have been more clear in my response: the data from my MY sent back to the Tesla super-computer in turn feeding the AI neural net with my drive data (and that of so many others) is training the auto-steer and FSD beta programs to become autonomous. What I see in the real world is that my MY driving performance steadily improves on very tricky roads. Amazing times we live in.
Yessssssss. Thanks gents.
The description of driving in SF sounds like about 50% of what happens in my town in the UK. I just hope I live long enough to see fsd Beta appear on my MY!
James please do not disregard the people in this ecosystem that feed your organic self. I get more than a couple potatoes from my farming efforts. I run a small farm and feed healthy food to friends, neighbors, and at farmers markets. Regenerative organic production systems are what I use. Healthy soil, healthy food, healthy people, clear thinkers like you. Appreciate your farmers who care about your life and health like I do, you are a treasure to us farmers who have more ecological knowledge and understand biological and biochemicals processes in carbon based systems. We rural folks are often confused by your silicon based world.
Peace out.
Z
Always a pleasure to hear James's clear and logical understandings.
It appears that 11.4.1 is still not choosing optimal lane choices early enough. Not sure if it needs better map data or road sign OCR.
Definitely seems smoother.
It just needs a better ability to think
@@fredpsimas1874 Interesting point and question.. How do we define 'thinking'? Perhaps the ability to reason? Would this not require an understanding of irrational human behaviour? It would have much to learn. :-) .
@@DoghouseFunkBlaster65 That's just probabilities and pattern matching. AI should be able to cope with that.
@@robertwhite3503 Indeed! The Myers-Briggs Type Indicator could easily be applied for pattern matching, but one would have to do a comprehensive analysis of each individual on the road. Only then would it be able to determine probable behaviour in a given situation - given the individual's temperament which also 'matures' with age and experience. If it can accomplish that, then our thoughts are no longer our own. ~ Scary!
@@DoghouseFunkBlaster65 LOL
Almost is never better than finishing In any situation 😊
THANKS,🤗 JUST WHAT THE DR. ORDERED 😎💚💚💚
Made circuit diagrams that way too. Etching circuit boards manually too. Fun times!
Wire wrap was your friend otherwise.
Thanks for sharing! Super interesting. I'm the 'Golden thread' guy and would love to see AI at work here to plot the perfect route guide for fsd with real-time traffic maps overlaying it's starting and destination. This 'thread' not only showing the correct lane but can also show things like speed, when to stat indicating, school zones and other important info.
My understanding of FSD is that it has no "memory" in terms of speed limits, construction, turn lanes, and the like. This is based on videos, podcasts, and comments i have seen wherein one finds a preponderance of reports and complaints to that effect.
It would seem an obvious move to integrate a Waze-style UI, which includes driver reporting/updating. That being said, i can't imagine Elon has overlooked this shortcoming. I am confident there is something in the pipeline.
Such a good and nuanced point. Even for the very concerned experts, possibly they see the probability of AI ending humanity as quite low, but even at quite low, it is > 0 and that is a really bad thing.
this is a very dour but honest prediction on FSD from james
Make it Weekly! Please 😁
man that was a great explanation .
I'm liking Bard better than GPT. Current data makes a big difference. Great talk.
I have a video of the cruise in Austin, there is definitely a autonomous car in Austin driving around 6th street area , I believe every evening , not in the mornings
Well that was a hoot
To amplify what James said... I was responsible for the monthly program at the RTP Home Computer Club in the early 70s. As a project, my 10 year old daughter and I designed and laid out the first floppy disk controller on our dining room table to interface with the S100 buss on the first production home PC (Altair 8000). We used a coated copper substrate and masking tape. This was sent to a fabricator who etched and drilled it for $15. The code to drive the 8" floppy controller was handwritten in 8080 machine code and "burned" into a 512 byte eprom.... times have really changed.
Minimal Lane Change is also good for chill drafting to make next charger 😊
I would love to see a tap to change lane for FSD. You tap on the screen to select the lane you want the car to move to when it's safe to do so. Tap again to cancel.
Hi, can you explain how fsd is going to deal with situation as police officer stop traffic or managing the traffic. I mean will it be able to read and react to hand signs for example?
My thoughts on AI effects on me are going to be great for me since I'm retired and looking forward to it
Afraid for my two boys in the thick of their working years. Excited for grandkids as they are a decade from college and growing up with it should be the best scenario .
Great conversation on AI and transitions!
Looking back at the agricultural transition (from everybody being farmers) led me to suspect it may have parallels to transition from ICE vehicles to EVs. I think of the iconic photo of farmers dumping milk because the price was too low. Agricultural surpluses led to price supports on one end and food stamps on the other. So, are we heading to a future where we subsidize ICE companies to produce EVs and subsidize consumers to buy EVs? Is that where we are already?
Yes, we are there... although it may be more analogous to the horse/automobile transition, with the opposite environmental/ecological effect.
Shouting this into the wind, would you be able to point in the direction of where to go if you wanted to refine a llm on a specific body of work or task?
Is this available on podcast platforms? We need a Dr. know it all podcast!!! Just convert the video to audio
Congrats you got to the shareholder meeting
The problem is FSD is not a well defined term and when you say it is almost ready, you have to say exactly what you mean by FSD.
Exactly! I want to barf every time I hear someone say "Solve FSD" in any context because it is meaningless. To James's credit at 5:05 he does mention "an interesting benchmark" of when _could_ Tesla take the liability once FSD is enabled. And suggested it might be in mid-2024. In my opinion, way too simplified and different than anything that will actually happen.
I wrote about this last year (Levels of Autonomous Driving is Stupid and Tesla FSD ≠ Autonomous Car, Are you being fooled?).
With regrds to your Hard Takeoff scenario, how can you be so certain about the predictiblity and difficulty of the emergence of volition, or goal seeking in AI when we don't really know how or when they arise in us?
Looking at the basics:
Ai improving itself needs many things but ultimately it needs to improve chips by itself to create future versions of itself with more computational power
John, you seem to be a very good hearted guy. Don't you think that a better name for your channel woulb be Dr. Who does not know it all?
For novels it should be fine if the context window contains the essential facts up until the current part of the story. Just tell it to make a summary and to add another chapter according to the things happened and a specified goal.
Another great point. AI kills us all, quite unlikely, but economic and social chaos, is what results in the end of humanity, somewhat likely. JD, is super smart.
Thanks. Mallicious users are already ahead of today's applied technology. What will they do (are they doing) with AI when it is advancing faster than it can be applied at scale.
Could there be a seti like distributed compute solution for the largest LLMs?
Google is operating in Gilbert Arizona and I have seen waymo vehicles going around and around downtown in Phoenix.
The future is a mushroom cloud! Science & Tech can't save us!
Are you getting any phantom braking issues. I drove down to Austin for the Investor day (10.69) and had several phantom braking issues when driving > 75mph. So far 11.3 seems better - but have not done a long trip yet.
I am reasonably well versed in computers and feel so far behind with how fast AI is developing
hopefully, you didn't buy shares before Elon started offloading his?
Rumor has it, we’ll have it in two weeks!🤣
I get that concept of the soft takeoff vs the hard. The product of likelihood and impact yields the urgency factor. The likelihood of something disruptive occurring in the soft takeoff region of the spectrum is high and the impact in terms of societal fabric is also high so the product of those two is large. Where the FUD prevails is in the hard takeoff in which the likelihood is low but the impact is extremely high, making the product potentially as large as in the soft takeoff case. But the likelihood factor is weighted more than the impact so the hard takeoff case appears as the worse of the two instead of allowing us to focus on the soft takeoff. Major disruption will happen in either case but we need to address the highest likelihood case to have any hope of mitigating the societal impact. Just my long-winded way of saying YIKES!
11.3.6 still gets wrong speed limits from a mysterious source and cannot read overhead LED speed limit signs. But even more annoying is that on the rare straight section of road the steering never stays still. It constantly moves slightly left then right and repeat. Only a couple degrees but I can make far fewer steering inputs on the same section. And WHY does it slow slightly for green lights at clear intersections when it's not following a car?
Quick nerdy question, when one inputs data into one of these avatar/agents, where and how secure is that data? Does it stay local somehow or the application vendor now know more about me ?
54:50 It says links in the description.
The links are not in the description. Just lots of ad referrals.
In Philadelphia, trucks and people just stop to deliver or run in to a store etc. maddening.
It is happening
US traffic is super easy. The real benchmark is the inner city of Rome, Bologna or Verona. After that the final frontiers: Delhi, Kinshasha.
Between the first FSD Beta release and now, have the tennants asked about isn't it about time to plaster the walls? Even the furnitures are on back order.
Would be cool if some did FSD parties, see how they do when everyone is also on FSD
Cruise also drives during non-busy hours
Pittsburgh pa is even harder for self driving/ new drivers.
I plan to replace my ICE Kia Sportage with a Tesla Y within the next two years. It would be great if I could pre-purchase Tesla's FSD Software pending the purchase of an actual Tesla Vehicle. Of course, Tesla would have to be willing to attach the software to the consumer rather than the Tesla Vehicle. Perhaps, it would benefit Tesla if FSD was offered to customers pre-purchase at a great discount! What do think Elon?
The big AI companies that are ahead will learn how to spoof the other AIs.
This about the product recomendation bot. It is drawing that info from somewhere and there is a massive insentive for companies to influence that data.
How will AIs be able to determine between real data and spoof data that has been set to manipulate the results?
As soon as the cars deliver themselves from the factory
What does "add another zero to the accuracy" mean exactly?
1:28. James talks about a statistical language response system. The context is actually a statistically mapped context. What it is not is a specific causal context. The big difference is causality v causal free single dimensional (language,text) state mapping without causal knowledge. BTW this is why OpenAI just stated that current neural model have come to their limits and new models need to be developed. Exactly the control problem with Tesla’s FSD. 1:32. Customization requires Small World models with specific causal context. You need Delta Learners for this which are different models.
Tax the bots
Driving too cautious is NEVER a bad thing ! If you are in a bus you don’t yell at the driver because he don’t go fast enough … so let FSD drive his way as long as : no security issue and no confort issue ! The way YOU drive is irrelevant (it is not about you!)
Ha, first printed circuit. I remember how exciting that was to me. Who knew for every chip and component ever made, you could get the design Fax free from the manufacturer (and a couple of chips to test with!). Assemble like Lego those on your computer drawing. From Kentucky, I could upload the file (by modem) to a company in Alberto, Canada, and by mail receive my boards in a week or so. Was so cool, such fun then. 🙂
I never used tape only 'pen' , don't get to talk jars CuSo4 laying about the place😵
I believe that at first, they will take the liability only for those with Tesla insurance.
James, I have a long running business bet, since 2017, that we will have 12 Mio fully autonomous cars in 2025. 🎉
My son in first year uni has expressed interest in becoming a fine work carpenter. I believe in advanced education but I also know it will be at least some time before robots and AI can do fine carpentry. They will eventually. Perhaps human made objects will be more valuable in future.
Transition to less work will be easy as an tsla shareholder :)
Holy cow 🤔
I like the analogy with new buildings going up and FSD/self driving. But it struck me that there won't be another tall building in S.F. for 10 years or until they fix the drug problem. Forget office space and who's gonna buy a million dollar condo if they get dumped out on the street with the homeless folks.
Does he work for Tesla FSD team
They covered this problem in Star Track don't give the emotion chip to Lore or Data.
The glue that holds it all together doesn't have to all be classic coding. Most of the constraints can be handled by the acceptable language grammar phrase applied to the situation 😊
I think the point Tesla takes liability will have a relationship with Tesla insurance. Customers will pitch in for the liability risk.
For me the issue of taking liability is obviously crucial but how will the car handle the handing back of liability? It surely can't do that while driving, i.e. it will have to be able to safely stop and not try to hand control to the human on the fly the way it can when constant monitoring is mandated as it is today.
It does change lanes from the fast lane once a car is approaching behind it.
In Europe such behaviour costs a ticket.
Gradually as more and more cars become autonomous this will affect the A.I. training suite as the A.I. wont have to cope with "crazy" drivers only crazy pedestrians, but maybe those can be "trained" with Neuralink🙃 .
They need to make it perfect, no intervening, no sudden jerking, no touching the wheel, no running into emergency vehicles on highways. Make it a perfect driver assist autopilot. This is best use case and gives it value as we wait for Robotaxi grade pilot.
There are people who believe that the AGI is better than humanity and deserves to rule the world regardless of the consequences. There are also people who believe that they can use AGI to become incredibly powerful, and that they don't care for any consequences. Therefore, it's almost inevitable that whatever materializes won't be in human control.
I've always felt the bar for FSD "winning" is when Tesla insurance will cover all accidents if FSD has been active for more than 10 seconds.
Maybe i should short the property/casualty companies like Mercury or Allstate