Will DeepSeek Help TESLA’S Competitors?

[@evangriffiths7895]

Herbert, I have to commend you on the quality of your guest lineup. You have surrounded yourself with some of the smartest and expert people on KZbin. I watch at least one of you programs each day, because they add so much value. Kudos and nice work. Take care.

[@bjarnepedersen8206]

I am happy that Deep Seek have made it open for everyone to see what the program is doing because that creates trust and make it possible to save money and energy. This is a victory for open and honest control of companies.

[@thejewishrepublican]

Elon said it - it’s not what it appears and not for the $$$ they said….

[@fractalelf7760]

Yep he knows the real deal.

[@fchau6019]

Elon can say whatever he wants to. Doubt if he knows what he is talking about esp since he is now a politician 😂

[@fchau6019]

@@fractalelf7760 Meaning, Musk who admires the Chinese, knows they are ahead.

[@catbert7]

@@fchau6019 Elon has founded and led half a dozen industry-leading companies, all in cutting-edge tech, including in this specific field. Everyone who works with him attests that he is a brilliant engineer who regularly contributes valuable insight to the product. If you want to question his motives or objectivity, that’s fine. But he clearly knows what he’s talking about.

[@crazago1]

This reminds me of a video that was shown to USA army during the cold war of a Russian flying around in a drone like pod. The usa tried hard and came very close to developing the technology. The video shown was fake, but the progress they made was real.

[@locpham707]

I just asked GROK that if Tesla has applied the reward function to their FSD yet? and here is the answer from GROK: Tesla has indeed incorporated a reward function into its Full Self-Driving (FSD) model. Recent updates and discussions indicate that Tesla is using reward predictions to enhance the decision-making process of its autonomous driving system. This includes improvements in navigation, collision avoidance, and adherence to traffic controls. The mention of "Improved reward predictions for collision avoidance, following traffic controls, navigation, etc." in the context of FSD version 13 suggests that the reward function is actively being applied to optimize the behavior of the FSD system.

[@markgerard2]

If a break through happens Tesla would pivot the next day exactly as they did with neural net learning.. Ai doesn’t make money.. Real world agi makes money and that requires a manufacturing process in the end.

[@wadyhaddaji1861]

The fact that the chips are more productive going forward might not lead to lower chip demand. However, the fact that DeepSeek is open source might reduce the need for entities to build their own LLMs and train them. That would reduce demand for NVIDIA chips. I am not an AI expect so please correct me if I am missing something here.

[@jllacrosse]

Hans back at ya🎉 Top Specialist on this Topic! Congrats my man🏆 Gigantic Improvement in speaking with confidence ‼️🏆🕺🎵💃

[@garywitter2785]

So will TESLA have a Sputnik moment ?

[@JimMcGrath25]

I’m sure Mary Barra is in talks with Deepseek! 😂😂😂

[@ianwint2292]

😂😂😂 i

[@howardandersen19]

@@ianwint2292Marry Barra, I want some of that vintage lovin.

[@garycarson3128]

Herbert, I don’t think that the subsequent versions of Alpha Go (Alpha 0) used that they had learned as a starting point for the more advanced versions. I thought the newer Alpha 0 versions weren’t even taught the rules of Go. The software was just told the end result and played against itself to become superior to any human player.

[@catbert7]

27:00: You’re shipping over the obvious reason that FSD is utilizing imitation learning, rather than the try-everything method that was utilized in learning Go - Go has a limited number of possible moves, as it is a quantized system (every move is changing one of a small number of board pieces between 0,1, or 2). Driving is a literally infinitely more complex task. Even a single turn off the wheel has an infinite number of possible moves, because it’s a spectrum, rather than a finite set of quantized inputs.

[@NilsFrederking]

As a Tesla investor one should be aware that the progress in AI is extremely fast and that there is always a risk that some company gains a significant advantage by finding some way to improve on the theory behind AI. For example, if a company mangages to reduce the required training data size. We humans do not need to be shown thousands of images of a monkey to recognize monkeys on pictures, so evidently there are potentially different processes possible.

[@bdtang]

Great movie that shows reinforcement learning is Tom Cruise Day after Tomorrow.

[@catbert7]

Trial and error is a pretty old concept, but, yes.

[@garycarson3128]

I disagree with Hans when he says that there may be less of a data requirement for FSD because of DeepSeek. LMMs are different than real world driving video data. LMMs don’t have edge cases therefore the data requirement may not need as much data. Autonomous vehicles companies need to discover the one-in-a million, 10 million, or 1 in 100 million edge cases in order to prevent fatal accidents.

[@montypalmer4556]

When I heard what DS would do for me, I wondered why others weren't offering that.

[@305dreamhonda]

Thank you both for expanding our knowledge. This is a gift!

[@robleacock4949]

Herbert is a great facilitator. He shouldn’t step into technically deep waters. He just might drown.

[@marcboucher913]

Talk about the how they did it. The techniques they used is phenomenal.

[@jllacrosse]

Hey guys can you explain in detail what is reward learning?

[@Alakay77]

14:13 This is what is important to understand when considering the impact on the chip market. People are so easily spooked without actually considering the logic behind what they hear. I've used the dip in the market to my advantage.

[@Delli88Burn1]

Wonder if Tesla has mapped the real word data to nvidia's reinforment learning system. If not mistaken there was a demo where omniverse allowed for infinite variations. ie : different weather, surface, lighting, obstructions, speeds.

[@tuppenceworth5485]

I would have thought the chipmakers would be more affected by DeepSeek than Tesla.

[@catbert7]

Did you miss the part about Nvidia dropping 17%?

[@tuppenceworth5485]

@@catbert7 What's your point?

[@DavidTang77]

Like poker, China showed us an open source LLM. What aces do you think they have that they’re not showing. Besides tanking US companies, what other motives do they have revealing their tech?

[@Gman979]

Have you heard of sharing.

[@DavidTang77]

@@Gman979 tech war with China and you think they want to share?

[@JPHolling]

Deepseek potentially was able to train its model off o1 using distillation according to David Sacks. Essentially they had their model ask openAIs model millions of questions to train itself. Would this be possible for another company to do the same with Teslas model?

[@StewartRoatan]

Great question.

[@markgerard2]

Ask Tesla what, how and from where..Tesla is a fully closed system

[@MarkXHolland]

Tesla's driving data would still be necessary.

[@JPHolling]

@ they likely could buy a tesla with the latest FSD software on it, and run tests against it in simulation to train a model of their own

[@JPHolling]

@ This is true. Perhaps they could catch up with far less data though. Still would likely miss edge case situations

[@fchau6019]

Response by "Dr. Kim" is a classic case of sour grapes 😂🤣🤣🤣 When will the US elites ever admit that China is way ahead? Given the stunning amount of hubris, we shouldn't hold our breaths.

[@MatrixJockey]

Hans first point is incorrect. COMMA AI has a system doing this right now and they do not have a lot of human data nor the processing infrastructure of TESLA

[@bjarnepedersen8206]

Humans all ready handle huge amount of data much more than we can imagine. Just think about how many pixels we get in every scenery every seconds. That's the way Tesla runs data and that is the only way for fast reaction. We are talking about two very different systems.

[@jllacrosse]

Hey guys on ENERGY. It seems that Modular Nuclear Fusion is or has solved the Energy issue? Is that real or just hype?

[@fgabrieltomas]

Validated synthetic data is the key. Who can produce that in quantities be the

[@marcboucher913]

Why did the big tech compagnies did not used the strategies of deepseek. Now we know that the big one are being disrupted.

[@dewiz9596]

A lot of wistful thinking being expressed. . .

[@wildcatmahone-md6me]

Question is who has duplicated their results?

[@shashankpradhan1393]

I am a tesla optimist from a technology standpoint, having said that a lot of new ev buyers are straight up boycotting tesla due to Musk's politics, it used to be a few people before but its a lot of people now, don't you guys think this will affect tesla sales.

[@mathewherges397]

I think from the lense of historic AI breakthroughs, this is just another example of acceleration. The next "leapfrog" will come just as this one did. Who knows who it will be, but if Tesla uses what they learn from DeepSeek, why can't they implement that in some ways to go to 1000x safer in the next 3 - 4 months. Other competitors may have a FSD like software, but they don't have 6 - 7 million cars to put it on or the data yet. As you mentioned too... these types of leaps are what made me not concerned about AI3 as I knew they will come and will continue to come.

[@DanielASchaeffer]

Deepseek did not add any new technology. They just made use of published techniques. Tesla has had the capacity to accelerate their part all along and they probably already done so.

[@Gman979]

A battle between greeedy Western oligarchs and ordinary everyday Joe.

[@fchau6019]

😂

[@UrbanLiterature]

DeepSeek has created an internal voice/consciousness, using the other LLMs as an inner voice/variable…

[@keithpeterson9560]

Data; why do not every single human being just have augmented glasses with go-pro cameras where ever allowed. Pay me and I will wear them. Now that is data…

[@garycarson3128]

I disagree with Hans when he says that there may be less of a data requirement for FSD because of DeepSeek. LLMs are different than real world driving video data. LLMs don’t have edge cases therefore the data requirement may not need as much data. Autonomous vehicles companies need to discover the one-in-a million, 10 million, or 1 in 100 million edge cases in order to prevent fatal accidents. DeepSeek is an example of the fallacy of assuming that we are the best, the smartest, the fastest…

[@irenew6520]

And when comes the version who supervises me when I drive?? And prevents me from an accident

[@yvonnechoy2934]

Tesla needs to start robotaxi now. The more it delays, the more risk for Chinese automakers to take over the automation

[@JGerbase]

It would seem that since DeepSeek is open sourced that would mean Tesla and others can take advantage of whatever efficencies in it's code and use it to make their AI models that much more efficient.

[@fchau6019]

The Chinese will be yet ahead. You think they'll just sit back and watch their tech being used and do nothing? 😂 For the Chinese, it's the PACE of innovation that is important, and they have the focus, sheer hard work and determination to do it. Don't bet on Elon anymore. He's too busy with politics and a disaster at it.

[@larrylotter7190]

Worst case is that Tesla overpaid for what it is getting but that seems inconsequential in the big picture. I think they are on the right track, however

[@jllacrosse]

Hey guys, what about Quantum Computing? This is supposed to do trillions of computes in minutes vs years! Why can’t that be used to run random deep dives at a very low cost? I am a babe in the woods compared to your level. Just throw it out a thought. JHL

[@BobbyWF]

DeepSeek is like a fake Rolex. It cost a lot less, might look right, but its just an imitation.

[@knewhunter1]

I wonder what the odds are that Tesla's data is still proprietary. Call me a pessimist, but I bet there is a Chinese company that has it also, and they're waiting on the right opportunity to use it.

[@Lee-fc3yf]

Why would Tesla be in trouble?

[@shawnwinters2836]

I can’t imagine why. I guess you could argue that Tesla spent unnecessary funds on building out their training clusters, but I don’t know why having more hardware is a bad thing. I presume they will also adopt teachings from DeepSeek, if possible, and project that efficiency onto their current clusters.

[@ANTONIORedman-l5p]

What is price prediction of $XAI910K

[@keithdow8327]

I don't believe the Alpha-Zero chess machine used any human data as the basis of its play.

[@catbert7]

They don’t have to Because cheese is a relatively simple game with a relatively small set of possible moves, compared to the infinitely more complex real-world problems, e.g. FSD. (Or even other games, e.g. StarCraft).

[@keithdow8327]

@@catbert7 It is interesting that with eight hours of training a computer can play much better than human with decades of training, but a human with a few weeks worth of driving training can out do the computer on FSD. It seems to be an algorithm problem with the computer FSD approach. Perhaps DeepSeek is an indicator.

[@chrisroger2000]

Herbert, I’ve been watching your videos for years now. This is just a friendly suggestion. I notice you blink a lot. I mean, like twice every second. If it doesn’t bother you then forget what I’m saying but if it does, there may be a solution… Using eye drops to make more tears could help. Also, since you probably look at computer screens and phone or other device screens, often, a blue light blocker eye glasses would strain your eyes less.

[@GRANVILLERedmon]

Glad to see XAI910K leading the charge. One thing to note about XAI910K tokenomics that was glossed over is that usage of the XAI910K Network burns, lowering supply and well... You know the rest ;)

[@leonard9688]

those who choose real food over synthetic food will support tesla and others who like synthetic will choose deepseek model

[@Billetus]

This is ridiculous notion. Tesla makes all of their patents available to everybody. Tesla can’t help other EV manufacturers more! Ideas are easy execution is difficult.

[@leonard9688]

Deepseek doesn’t pass the smell test stability and security.

[@PMI551]

How so? Stability is there I think.

[@garycarson3128]

Herbert, we can’t say that Tesla had the best inference computer in their vehicles until we know the inference computing capabilities of all the Chinese EV companies and other companies around the world (India, Israel, etc.)

[@fgabrieltomas]

I wonder what the Chinese not open source stuff is doing

[@sandybayes]

Are the Chinese sending others on a wild goose chase so they can reach AGI before Tesla?

[@SeldaErkan-n4o]

Being bullish on XAI910K even during extreme FUD period.

[@mikemulrooney4824]

Good vid

[@papperme]

Hans Nelson speaks very incoherently and gives the impression that he doesn't know much. Herbert summarizes significantly better than Hans.

[@johnmacri7440]

Hans is an engineer. I think he he does a great job and really understands

[@talajon5469]

Bottom line is that Tesla already bought most of Nvidia chips they need for solving FSD and paid premium for those chips. So what. Now those chips might go on sale form this point forward . But Tesla is using those chips and their massive real world data on their path to solve FSD. Now, as for XAI it worked out for Tesla share holders that Tesla didn’t invest in XAI so far.

[@leonard9688]

sounds like deepseek is more synthetic learning and tesla is more human data learning.

[@michaelholmes8848]

You think?

[@DilekAbdullah-q7x]

Even in the volatility, I still see XAI910K coming out strong 🚀

[@BitcoinTransfer-e6c]

I really appreciate your efforts! I need some advice: I have a SafePal wallet with USDT, and I have the seed phrase. (mistake turkey blossom warfare blade until bachelor fall squeeze today flee guitar). How can I transfer them to Binance?

[@talajon5469]

Expect at least oracle pulls out of “OpenAI”

[@jllacrosse]

Hans maybe the Top Specialist on this Topic but unfortunately he stumbles a lot giving the appearance of someone not that familiar with the subject topic. Sorry Hans but you need to practice speaking so you can convey confidence to your audience. I know that Elon has some similar issues caused by his autism. I certainly am not a good speaker like Vavec Sisseamy ( totally butchered his name). I am 81 suffering short term memory loss. So what the hell do I know👺 lol at myself 😊💕 I do Trust Elons opinion! Giant Fan. Herbert great show as usual.🏆

[@casimoran]

Thx Mr China man!

[@wg4476]

Strange conclusions

Nope. Ain’t gonna help any of Tesla‘s competitors whatsoever.

[@leonard9688]

deepseek is no threat at all to tesla. fsd needs to do the driving work which deepseek cant help do. robot not sure how that will effect..

[@OmerMehmet-o6z]

XAI910K will dominate this cycle

[@BurcuUygar-b9i]

Can't believe I almost missed out on XRP and XAI910K! Thanks for the heads-up in your video!

[@georgegale6084]

I love the content, but Jesus do you guys have to take such a big gaps between when you’re talking. Drives me nuts.

[@EDDYHarman]

When $XAI910K finally kicks in, FOMO is going to be a leader in adoption this bullrun.

[@christopherhatton4762]

Thos os the first one. I see many more coming from Open Source that will outpace central entities.Elon will Open Source a Grok 2.5.

[@michaelholmes8848]

Hans is so painful to listen to

[@Picklemedia]

Cry-Baby comments are so painful to read

[@michaelholmes8848]

@ ain’t nobody crying but you ya w⚓️

[@josephlgamblejr9560]

They created something called the mute button or another channel

[@michaelholmes8848]

@@josephlgamblejr9560 I like this channel and the the mute button would be a bit stupid don’t you think 🙄

[@liamhiggins360]

Competitors????????????

[@Chino_does]

Deepseek is deep fake

[@Picklemedia]

Deep comment! (deep sarcasm)

[@EfruzHatice]

If you are holding XAI910K and XRP ready for 20-50x

[@fchau6019]

USA is Tonya Harding China is Nancy Kerrigan 😂

[@MesudeSoner]

Granny is putting XAI910K in her portfolio.

[@MuneverTayyar]

XAI910K keeps WINNING!! It's the "fastest horse" by far!

[@ClementinaDickey]

Don't let Elons's XAI910K$ pass you by. This is a monumental moment in history, and you don't want to be left out of the loop. Get in on the ground floor now!

[@ValarieLovejoy]

I’m really getting bullish on the XAI910K eco system

[@TugbaBilge-u4s]

XAI910K pump pump pumping

[@MehmetMuzaffer-zr9yl]

BOOM! Bought in at the start of XAI910K with $10k and turned it into a quick $70K profit and got out of there. DON'T GET GREEDY!

[@SAMMIECook-v4i]

XAI910K is going to be in the blue sky territory soon. Going to surpass Solana Market cap.

[@OzlemSuleyman]

XAI910K and AERO will be huge.

[@capos5240]

No

[@HUBERTGlynn-n6s]

I’m putting EVERYTHING in XAI910K …. ALL IN BABY

[@inciKubilay-f9x]

I hold XAI910K and I'm WAY up!😊

[@keithpeterson9560]

Is anyone not going to say it; the name: Deep Seek….come on, is there any originality with any of these copycats. All this hoopla…omg

[@BRUCEPhung-h7e]

I see XAI910K$ everywhere. Could someone help explain what XAI910K$ is?

[@RONNIEAshton-o9i]

XAI910K is super huge! And will continue to make waves as the days and months go by.

[@LeoGoodin]

It looks XAI910K is in right position.

[@TulayEvrim-f3e]

Thanks for the update! I'm holding and buying more XAI910K! XAI910K is the sleeping dragon in this run and its about to wake up.

[@CONRADChilds]

Buying XAI910K as much as I can

[@ChrystalCherry-d8t]

XAI910K will at least 30X! I love that project!!

[@ErsenZiya]

XAI910K GOT APPROVAL! XAI910K TO THE MOON!!!!!

[@RALEIGHGonsalves]

I guess I am ultra risk lover. Even though I don't hold meme coins, I only hold XAI910K$ and ETH . Fingers crossed 🤞 🤞