Inference: Why Tesla will win AI 🤖
HTML-код
- Опубликовано: 2 май 2024
- On the Q1 2024 conference all Elon Musk address how Tesla's advantage in data inference was a key advantage in self-driving technology. Tesla's in-house designed chip is incredibly energy efficient and allows the cars to make split second decisions on the road. This same technology will be repurposed to make the Tesla robot think in real-time. This is a gamechanger! I was listening to the BG2 podcast while Supercharging and had this thought! Let me know your opinion in the comments below!!
My X: / gfilche
HyperChange Patreon :) / hyperchange
Disclaimer: I'm long Tesla stock nothing I say is financial advice. 
Развлечения
Hilton will install up to 20,000 EV chargers at 2,000 of its hotels across North America
Need to pay for parking in addition to paying for the charging
Inference is important but you missed a huge slice of how it can be used. It's compute. Compute that is never fully utilized. It's a bit like the seasonak spare capacity that sparked Amazon web services, that became a huge slice of Amazon's profits. Tesla vehicles compute, if owners opt in, can be used for a bunch of useful things. Including the tesla FSD model optimization. As well as research like climate models, cancer research, maybe even block chain stuff. The pint is It's another valuable resource.
@@simonmanning1844SETI@home
Nice explanation! Thanks.
100% spot on.
1:09 There's a glitch in the matrix.
Love your insights!
"Its brain will be where its brain is" - Galileo Russell, 2024.
How is the Echo chamber going?
Did you just learn about word inference recently
ya cant u tell lol
Great video, I learned stuff and it didn't take long.
Thankyou Gali, the penny drops!
Hmmm....this is no new information Gali....😂 But good that you are addressing it
Thanks for explaining inference.
He explained it poorly though. We distinguish two phases: Inference is just when a neural net makes a prediction. The other phase is training.
Distributed inference compute as a service TAM is massive if Tesla can get it to work. Coordination and networking limitations are key issues but maybe Starlink can provide a private network infrastructure needed?!?! 🤔
Where did I here tesla working on its own wireless connection faster than Bluetooth? 🤔 I’m sure it was not long ago..
Hence the reason for the jobs related to 6G telecom…between cars.
Inference computers will be in every Optimus as well. This is how Tesla will create economies of scale for the company in inference chips.
I'll Try FSD again when they get to Version 14. What's wild to me is that Elon thought they were close 5+ years ago!
So v12.5 😂
5 years ago is not that long ago in reality
Definitely another 5 years to go. We will hear end of 2025, then 2027. But it all ends 2029, that's when the next breakthrough will arrive, and deliver FSD.
Just to clarify, inference is the ability to take in unique information and solve that unique scenario. Tesla is innovative because they are deploying decentralized models while training on those decisions after the fact using one centralized model.
Gali is awesome
@hperchange would you explain why Elon should get the pay package based on the financials of Tesla?
Interesting thesis. Sounds true. Meta has 300k GPUs but if you count the Tesla hardware 2+ inference engines, Tesla has millions of GPUs.
How does 800m from Illinois change the Rivian outlook? Haven't found the law or any more info but can they get another 800 next year?
They will also use a double AI model system to do compression and decompression of information to simplify data packets and making Split Second over-the-air decisions possible in conjunction with the starlink constellation
Get super impressed about words about products. But I’m going to reserve bullishness for actual tangible objective results. Humanoid bot is a big fat zero to me at this time.
My recent FSD has critical flaw. Have Tesla contact me!
I thought tesla solved fsd back in 2020. What are you guys saying. We dont have fsd yet?
Inference workloads contributed to roughly 40% of NVidia’s datacenter revenue in 2023 or $7.4 billion for the last quarter.
Now imagine potential revenue of inference computing using the 5+ million inference chips of Tesla’s fleet
I can’t 😐 can you just explain it 🙏
Break it down to FLOPS if you want a meaningful comparison. How many FLOPS of inference compute did NVidia sell last quarter, and how many FLOPS are on board the vehicles Tesla sold last quarter? I’m guessing that NVidia is orders of magnitude ahead in that comparison.
What means the word „inference“, or: what is an inference chip (what does it)?
Inference comes after training. In the training phase, a machine learning model is trained based on historical data. In the inference phase, the AI takes actions on new/live data.
@@stefanmei5749thank you.
Yall becoming too bull on Tesla year ago yall were hyping up robotaxi etc
This is Gali at his f***’n best !!
I don’t know any computers that are faster than a human at deciding stuff while traveling at 75 mph. I’ve test driven 4 teslas and sales people never show me FSD.
Define win AI
Wow, that was elegantly stated. I didn’t not grasp that until now. Can you steel man that? What are the possible push backs? Does NVDA make inferences chips?
Our reality simulation is run on local inference too. We are projected with the resultant localized outcome to our senses.😂❤👍
I think this is clearly case show the meaning of Visionary. Elon start doing AI a decade ago. Everything Tesla led, everyone else followed
Remember 5G bro? It is the next great thing 2 things ago. It must be now...
I'm still shocked by how many people don't understand the amount of compute that is required for AGI or even close to AGI capabilities. The chips that are in our phones, our cars..etc are not there yet (not even close). So when Elon is saying Tesla car's from 2019 will be able to be 'full self-driving' I find it misleading. The best case in the next 5 years is AMAZING assistance driving...but your car won't be able to handle situations it's not seen before and requires human level reasoning!!
Now you're wondering what is required to reach that level of compute required...I COMPLETE PARADIGM SHIFT. It would need to be as big or bigger of a shift than going from vacuum tubes to semiconductors.
I love Tesla products, I own tesla shares, I love what Elon is doing with SpaceX...but where we are in AI is far fetch.
Just my opinion..don't sue me. lol
They will just layoff their AI team next lol
Inference happens no matter what in any AI use case… the significant thing you’re pointing out is Tesla’s ability to perform local inference, as opposed to doing it in the cloud.
Apple is also deep into the local inference game. Modern iPhones are incredibly powerful, including for inference. But Apple (so far) has shied away from competing in the AV space, so Tesla is alone.
Thank you. Apple just got out if that race. They can't compete in cars in my pov.
Nobody want lame latency in a robot.
Nvidia sells an inference chip for cars which is just as good as teslas? You can literally buy this tech off the shelf from them.
Just say inference in every sentence when you talk about AI 😂
Russell
Buy Tesla & Nvidia Stock NOW
get help
Typical case of conformation bias.
Typical case of nerd alert
Bullshit
That was a rambling stream of nonsense. What evidence does this guy have that Tesla are ahead? It’s just the usual Tesla fanboy wishful thinking.
You must be fun at parties
These videos don't rope in retail anymore. They got too smart. Tesla is done. Jump on the Amazon train.