Recent content by alee

I seriously doubt emailing adds any value to the feedback loop. Explaining what didn't happen, should have happened, etc. is not something a console operator can really influence, esp. if we are talking about collecting mass video training sets. Asking a human to sit in the middle to moderate...

At some point you'll either have to update or have your car neutered. https://electrek.co/2020/02/14/tesla-force-software-updates-owners-resisting-them/ FSD beta is a different beast than the main line builds. There is nothing but upside for 2021.40.x on the non-beta side.

With training sets, as long as they can be classified, more is always better (at this stage). We are extremely early days here... regular AP is based on billions of miles driven. I doubt they have more than 100MM miles of FSD on the books. For instance, they definitely don't have enough capture...

A big piece of driving is about showing intent through action. We drive in a way that says: "I intend to make this turn (and you should yield to me)", or "I am not moving (so I am yielding to you)". Accidents happen when that predictability goes away. We get frustrated at drivers that are doing...

They need a lot more drivers and data to be successful. I can't see a scenario where the beta would be too big at this point ... only too big, too soon.

65-30 would be unusual. I've seen it scrub 10-15mph at highway speeds (e.g. 60+ mph), more akin to you suddenly lifting off the accelerator. It feels more severe because you're not anticipating it (same reason passengers tend to be more susceptible to getting car sick while you're fine as a...

I believe the "request FSD" button came with 2021.32.22. So until you're at a current, non-factory production build (likely 40.6), you won't see it. Once you do, then you can opt-in and play the Safety Score game. You cannot request an update to a newer version - you have to wait until Tesla...

I'm certainly one of the ones who got upsold too hard on the good. My first 1000 ft on day 1 was the worst, but it definitely set appropriate expectations and spooked me good. I think had it been smooth and incident-free, I might have become complacent too soon, because sometimes it's actually...

That's what I would expect... if 2 nodes cannot agree, the tie-breaker should be to abort. It's definitely the safest outcome.

I agree with this long term, but it assumes near and far depth are able to be computed well by vision. Karparthy demonstrates some very nice depth mapping from a car's immediate surroundings, but I remain unconvinced for long range, where there is only 1 camera, which previously could lean in on...

Thanks... this is interesting to me. Typically you see redundant architectures work in parallel and sometimes by consensus. You see that esp. at the chip level as EPEs become more relevant with chip densities increase. Also seems odd that the B-side is asleep until the A side fails, esp. since A...

Lidar and radar is superior to vision-only for obstacle detection in 2021. Prove me wrong.

Not comparing who's ahead or not. Just pointing out that it clearly works. Only the Tesla religion says radar and lidar are wrong. The reality is radar and lidar are both perfectly serviceable for certain purposes. For stuff like phantom braking, lidar > radar > vision all the time. Using 2 or...

Unlikely unless they can trust the interior camera. 10.5 was the first pass at this, so probably a few iterations to go.

I did Waymo in AZ a few months back. It uses radar and lidar. It works. It doesn't solve Tesla's problem, but it does solve a real-world autonomy problem very well, without a driver. You can drive a Tesla everywhere in the US with FSD beta, but always with a driver.