Ashok Elluswamy, VP of AI at @Tesla on self-driving: "It's so obvious you can solve this with cameras. Why wouldn't you solve with cameras? It's 2026. The self-driving problem is not a sensor problem, it's an AI problem. The cameras have enough information already. It's a problem of extracting the information, which is an AI problem." (via @aelluswamy's presentation at the 2026 ScaledML Conference on January 29th)

@SawyerMerritt @Tesla

@SawyerMerritt @Tesla If I hear another idiot doubting vision for driving, I’m going to scream.

@SawyerMerritt @Tesla 🚀🚀🚀

@SawyerMerritt @Tesla I did a quick--well actually rather long lol--explainer video on Ashok's talk for those who are interested: youtube.com/watch?v=cEQYA9…

@SawyerMerritt @Tesla Notice he said “can solve” not “has been solved” It’s always a “next year” for Tesla

@SawyerMerritt @Tesla 1. Your head can move. Cameras cannot. 2. It's not a sensor problem it's an AI problem means they must suck at AI because other people have solved this faster than them with other sensors.

@SawyerMerritt @Tesla Vision-only was always a compute problem, not a data problem. It's just taken this long for the software to catch up.

@SawyerMerritt @Tesla turns out 'more sensors' was just a cope for 'not enough compute'. we've had the photons for a decade, we just finally got the weights right

@SawyerMerritt @Tesla How do you drive under heavy fog/rain/snow conditions?

@SawyerMerritt @Tesla More sensor input, safer system... I assume no one argues on this point

@SawyerMerritt @Tesla Cameras work. Trust the matrix.

@SawyerMerritt @Tesla Makes sense, seems like AI software is finally the real bottleneck, not the cameras themselves

@SawyerMerritt @Tesla Honestly I got a little scared when I first glanced at your post and saw Ashok's name and title! 😂

@SawyerMerritt @Tesla Haha @aelluswamy is almost upset now

@SawyerMerritt @Tesla If Two eyes on a swivel can drive a car, a 360 camera with depth can as well with the right AI models.

@SawyerMerritt @Tesla Non camera sensors solutions are like slow death drugs....

@SawyerMerritt @Tesla Yeah, but has Ashok checked with all of the keyboard experts on here before making such claims?? 🫢

@SawyerMerritt @Tesla I'm just here to say that if you're the VP of AI at Tesla, you know things that most people don't.

@SawyerMerritt @Tesla Literally drives me everywhere!

@SawyerMerritt @Tesla He keeps doubling down on this, yet the tech still doesnt work after 10+ years of development. It still requires interventions every couple hundred miles, and thus requires strict human supervision at all times, making it worthless.

@SawyerMerritt @Tesla If the cameras had enough information already in HW3, why did they increase the camera resolution for HW4? (Pssss, I know the answer).

@SawyerMerritt @Tesla 🤣 how did you get here today? In a Waymo

He is incorrect and not smart as he thinks he is. Humans drivers use Sight, sure, but also importantly Hearing and Proprioception. Humans eyes are also NOT fixed cameras, but complex computational optical sensors with multiple zones with significant parallel processing integral and also the use of very high performance muscle powered gimbal system.

@SawyerMerritt @Tesla Then why do waymos have robotaxis operating and Tesla doesn’t I mean if it’s so obvious?

@SawyerMerritt @Tesla It is literally to the point that I use my Cybertruck constantly but I never drive it. The only time I touch the wheel is to park it at a location since I can’t tell it exactly where I want it to park. The safety of letting it drive is way better than when I was driving it.

@SawyerMerritt @Tesla This is the way @Tesla_AI

@SawyerMerritt @Tesla Ashok is telling the industry the cold hard truth. You’re wasting your time with all these other sensors. It’s an AI problem.

@SawyerMerritt @Tesla No man. It's physics. Why not just use better sensor that can see further so you can have more time for compute?

@SawyerMerritt @Tesla The fact that this is debated today simply comes down to the fact that detractors just don't think ai will be good enough to extract the relevant information that humans are able to Thus they conclude it must be solved through sensors

@SawyerMerritt @Tesla One way to prove this to yourself is if you had the same 8 camera streaming into a human brain, (VR headset, zero lag etc) can they safely drive the car. If they can then it surely is an AI problem, nothing more. This is proven to be the case day after day.

@SawyerMerritt @Tesla It's even more obvious that camera washers are required

@SawyerMerritt @Tesla If Ashok is not working 24/7 at Tesla AI HQ, and doing a talk instead. That means FSD is solved and done. So he’s chilling

@SawyerMerritt @Tesla His answer seems so obvious even to a moron like me, but the insight to get from where they were 10 years ago to here is the difference between success, scalability, and failure.

@SawyerMerritt @Tesla He’s just regurgitating what Elon says. Not using radar is a huge miscalculation given the cost and roi.

@SawyerMerritt @Tesla Humans use their eyes to drive and they crash all the time.

@SawyerMerritt @Tesla Let me try solving it on Indian roads! Hr will need different AI model in each city and model will be so big that its hardware would occupy the boot of the car