Shad Torrie

@thsottiaux Merge live voice conversations with Codex!! Have it be an assistant sitting on top of the jobs running. I can check statuses, start new tasks, brainstorm ideas, and maybe even get pings when jobs are complete, all with voice. Like a coworker sitting next to me.

Now that we merged ChatGPT and Codex, what should we merge next? What's the double or nothing move.

@ZacksJerryRig This is what all other self driving car companies do. Including Waymo. In fact waymo is more dependent on mapping than Tesla. So I guess they are "level 2 forever" as well.

Level 2 Forever.

@ZacksJerryRig @xdNiBoR @Tesla Where is your source? Robottaxi tracker seems to see them very frequently and that's only the rides that Tesla enthusiasts log to it. Estimates say only about 5% to 10% of rides are logged to the tracker.

@xdNiBoR on any given day @Tesla has maybe 0, 1, 2 or 3 unsupervised cars on the road (1 year after 'launch' 😂) if you think tesla doesnt have a team of people watching each of those single digit cars like a hawk then I have a bridge to sell you.

A bit disturbing how JRE just seems to completely ignore actual unsupervised rides? He can't be that dumb right?

@thsottiaux I submitted this feature request too late, but maybe next week??

Lots of improvements across the board both for coders and general work. Update your codex app to enjoy all the improvements. We will be back next week

@wholemars @DevinOlsenn Haha everyday! I'm just saying that the potential for objects directly in front of the car is why Teslas backup first before driving forward.

@ShadTorrie @DevinOlsenn Do you often get out to check if there’s someone lying in front of your car?

FSD struggles to get out of parking spot due to lack of front bumper camera. This is a clip from my latest YouTube video, but it felt worthwhile to post separately and discuss. This to me is a clear example of FSD getting itself into a situation that it can’t get out of due to a lack of a front bumper camera. You can see the car wants to back up to gain vision, but becuase the spot was so tight it couldn’t back up anymore. What do you all think?

@thsottiaux multiple account support for easily switching between my work account and my personal account. Can't bill my company for my personal vibe coding.

Send us feature requests for codex in the form of an images 2.0 generated image. It makes it easier for codex to implement if we decide to go for it. Saw some good ones today already that codex is cooking on.

@wholemars @DevinOlsenn A human could get out of the car and check to make sure there wasn't a person laying in front of it. Or they would see whatever they need to as they are climbing into the car.

@DevinOlsenn Would a human also be unable to get out of this spot without a front bumper camera?

@thsottiaux when I hit my rate limit, it would be really nice to schedule a task so it will run when the rate limits reset, also during low usage hours could be useful if you release a save on usage during off peak hours.

@wholemars @SawyerMerritt @DirtyTesLa Who needs lidar when you have pop ups??

@SawyerMerritt @DirtyTesLa pop ups will solve everything

Critical means you might have died if you didn’t take over. Potholes would probably be “discomfort” unless it was a massive one

@wholemars Also this data can be used for what is called self-training. This has been shown to be highly beneficial to deep learning models to increase dataset size and quality.

@wholemars They should and definitely are using the FSD miles for training as well. Because that gives them exact feedback on what it is doing wrong. Using the real-world user corrections as reinforcement learning feedback.

This is based on a common misunderstanding of Musk’s tweet. He said, in discussing Nvidia and how hard it is to build a self-driving model that you need 10 billion miles of training data for unsupervised. People misinterpreted this to mean you need 10 billion miles of FSD usage to achieve unsupervised. But FSD usage is not what the training set is composed of. The training set is made up of *manually driven miles*. Human driven miles. How is the model going to learn driving by watching itself drive? It needs to watch humans. Tesla has been able to put together 10 billion mile+ training sets for some time. That’s why they have already been able to deploy unsupervised vehicles. Nothing magical is going to happen when FSD passes 10 billion miles of usage, other than it being a cool moment. Of course, everyone who misunderstood the tweet will probably start screaming about how “Elon lied!” when unsupervised doesn’t magically ship at 10B miles of usage.

Here's my question to those who have used unsupervised in Austin and FSD 14.3. Is it apparent that unsupervised is on 14.3? Any obvious signs that is or is not?? @DavidMoss @DirtyTesLa @AIDRIVR @tesla2moon

@nathie @noridoesvr I was so excited to see where they would take the hyperscapes. This is the only time I ever used a horizon world.

@noridoesvr No hyperscape is a huge blow

Meta is set off to kill Horizon Worlds (including Arena/Central) for Meta Quest headsets, along with the ability to view Hyperscape captures within Worlds.

@Teslarati 10 days is just too short to gauge market interest and demand. Especially for this big of a purchase.

🚨 We’ve seen some Tesla fans talking… What do you think about this whole “Just for 10 days” situation with the Cybertruck AWD?

I often find myself brainstorming features with ChatGPT and then going to Codex and having to redescribe the feature. It'd be awesome if you could connect them. So I could say to ChatGPT "/codex go do that". Or somehow link a chatGPT thread to Codex. @thsottiaux @sama

@wholemars Not if you're on windows... 😞

Loving the new Codex Mac App. One of OpenAI’s strength is building great native apps.

@Teslarati For long road trips, they could also add the option to move to a fully charged cybercab rather than wait for the old one to charge. You might not want to with luggage, kids etc. but helps those that complain about charging time compared to filling up gas.

It seems more like a last-ditch option where inductive charging isn’t available. Much like charging for trips, stops are only a few minutes long, so when Cybercab is operating, inductive charging will only take a few minutes. This traditional plug-in route is likely for when the Cybercab arrives back at its hub and is being cleaned and serviced. It’s likely quicker and will get the vehicle back out on the road within an hour.

@DavidMoss @robotaxi @ZacksJerryRig you're being mysteriously quiet on this one. 🤔

Unsupervised without chase tailing cars are truly here in Austin!!! 🙌 Ride 58 was my magic number, what a truly awesome experience. Made the whole week worth it & then some. Tesla Robotaxi Model Y flys under the radar, nobody knew I was in a self driving car. @robotaxi

@Skate_a_book @iliketeslas What about short flights within the US? Fall asleep at night in your car and wake up half way across the country! That's where I would see the savings.

@iliketeslas Because that’s your situation means it’s the same for everyone else huh. I can confidently say of the 24 years I’ve been driving my all-in parking fees is nowhere near $300

People complaining about the price increase of FSD don't understand the value difference between supervised and unsupervised FSD . I can see unsupervised saving me around 300 $ a month just in parking fees in Miami never mind insurance savings and more