Loading video...
Video Failed to Load
Tesla just rewrote its compiler and runtime from the ground up using MLIR for FSD with v14.3 and I'm guessing ~90% of you can't explain what that means. Since this video received much praise, I'm sharing it here. Even if you've already seen it, don't be afraid to share... show more
106,694 views • 5 months ago •via X (Twitter)
39 Comments

Fantastic explanation! Thank You Dillon.

i love this stuff! my pleasure

lol. Y’all made a little money making YouTube videos during covid and now you act like your experts on all things AI lol

or im just a guy trying to learn for himself and sharing what i find if you watch, i literally say im not a ml engineer or expert

Thanks dude, truly appreciate your work.

you kno its my pleasure appreciate you tuning in 👊🏻

So with every new release now we should see fewer annoying older habits come back?

potentially but no guarantees thats how it plays out, depends what tesla decides to do on infra side

@Xil_llix @grok, why is it better

@JoeTegtmeyer One of the most important attributes of Chris Lattner that you didn’t mention is that he’s an Apple Alumni and inventor of the Swift language and LLVM. He’s one of the top compiler guys on the planet.

People are most frustrated about the constant delays, lack of communication, and missed deadlines.

Excellent job on the breakdown Dillon👏

Makes sense Dillon. Super informative. Thanks for posting. Yes please do more.

Doesn't matter you can break it down in the most eloquent of ways but at the end of the day this puts a major delay in unsupervised and likely cybercab rollout another 6+ months if we're lucky.

你可以看看我写的两个线程。

What it means, like pretty much everything Tesla, they make some progress on some tech, like Dojo, Optimus or FSD, don't meet Pedolf's fantasy hype, he fires most of the team or they leave because of his bullying, and new guys are ordered to start again. Complete clusterfuck.😂🤣

💯👍😃. Great research. Thanks Dillon 🙏

youre welcome brotha!

💯👍😎

@Xil_llix This version, 14.3, will not be deployed to and does not run on Hardware 3 (HW3) vehicles…

Great job Dillon.

Can we just fix the damn nav. Literally the thing average users care about.

Dillon, the reason they adopted MLIR is that HW4 has a memory problem. June 2025 "Have to be super frugal with memory bandwidth", Jan 2026 "AI5 is existential". V14 is still small model because large model doesn't fit. V15 will require AI5 in 2027 and will delay Robotaxi

Loved the video. Nice understandable explanation.

Very helpful video, thanks Dillon, great work!

Thanks for sharing Dillon 😊

Thanks for taking the time to explain it.

@grok, what is MLIR?

This is great speculation. If the compiler was the "big feature", Tesla could have communicated that. There is fatigue from, "[the next model] will be the big break". Too many missed release estimates. More clear communication would thwart the negative sentiments.

Great video!

Great job with the breakdown, thank you. BTW the video should have been 3 seconds longer.. 😉

Very cool stuff Dillon! Many thanks for the deep dive and sharing.

고마워

Thank you

Epic find!

@TeslaBoomerMama Good episode Dillon

@JoeTegtmeyer Great breakdown of v14.3.

What’s your proof to this claim?

Thank you for posting video directly on X
