GoreTex is a bad example - it's gonna delaminate after a year or so of heavy use and is pretty much impossible to repair after that. Which also undercuts ACRONYM's messaging about their GoreTex products being some kind of like, buy-it-for-life rainjacket.
I’m sure there’ll be a PS6, but it honestly seems like non-portable consoles are on their way out. Wouldn’t be surprised if there’s no next-gen TV console XBox at all, or if it’s just a MS branded gaming PC.
From someone that worked in the industry ~6 years ago, it's clearly going well for them - frankly, they're expanding and scaling way faster than I would have thought possible in 2019. They've got something like 6 cities running right now and what, 3-4 more announced?
Another thing to keep in mind is that rideshare revenue in the US is extremely geographically concentrated in urban cores. This is why every AV company was targeting SF as their first city (excepting Waymo, which did some stuff in PHX). 'Hyperfocused expansion' probably looks a lot closer to tackling new, novel areas in different metro areas rather than, say, expanding down in to San Jose and the central valley.
These things, they take time.
They've clearly hit (or projections confidently show they'll hit) a point where each car is profitable. I worked in the space for a while - platform upgrades (new cars, sensors, etc) are planned out years in advance and are pretty complex processes. But generally, each upgrade was a massive decrease in cost per car. (usually 50% cheaper or more). So also possible they want to wait for the next platform transition.
Afaict dojo is just a training asic integrated into a cluster, just like other big tech companies have had for years. If anything at all it’ll just mean a lower cloud or compute bill for Tesla. It’s telling that Elon is the one that likes to mention Dojo and Karpathy is generally quiet about it - it’s not actually a quantum leap in ML capabilities.
They’re not tackling it with an end to end model - all of the ML at Tesla is used for perception (and some prediction - I think there’s a cut-in prediction head running), which then feeds into traditional robotics algorithms, like I described in the parent comment.
If you listen to any of AK’s talks he only really talks about perception and he’s basically said that right now none of their planning is based on ML models. This is backed up by what reverse engineers have found in their Tesla’s.
The “fleet learning” pipeline Elon described at Autonomy Day doesn’t actually exist in practice. They have some hardcoded heuristic triggers that capture snippets of data to be annotated by human labelers to train perception models off of. The bottleneck is labeling/engineering.
Autonomous vehicles by and large aren’t a data problem anyways - it’s a robotics problem conditioned on the output of ML models for perception and prediction. The majority of the work is just implementing all of the bizarre edge cases inside of the robotics stack.
The idea of tackling this with an e2e ML model is a pipe dream touted by people that aren’t familiar with the space or trying to hype up their approach. Whenever this approach is attempted teams will very quickly realize how untenable it is and return to implementing individual robotics modules.
tl;dr there’s nothing special about Tesla’s approach to AVs or machine learning and you shouldn’t expect them to leapfrog the competition because of some nebulous ML-based advantage.
Coming from someone who has seen the interiors of both the FSD beta and the stack at an actual AV company.
Dumb question - does the way the cannabis was consumed affect the accuracy of the device? Does it work as well for edible cannabis as it does for a joint?