Statistisos

@andrew_n_carr Holy residuals.

somebody made a huggingface model visualizer!! just plug in the url and explore at any granularity

@techdevnotes Better than 4.20's pricing, but it's sad to see the pivot away from the low prices that 4.1 Fast had.

Grok 4.3 has 1M context window with pricing of $1.25 input per million tokens

@scaling01 Yippie, now my limits will last 2 prompts instead of 1.

2X usage limits for the next 2 weeks for all Anthropic users on weekends and outside of rush hours on weekdays

nvm it works now 🙂

Grok 4.20 Beta 2 still can't read files attached to Projects 🙃

xAI needs to Fix the issue where Grok 4.20 model cannot Read the Files when uploaded to Projects

@techdevnotes 12 more hours in CA 🙏

The week’s over Grok 4.20 Beta 2 is still pending

@TailosiveEV @djcre8or Lidar directly measures distance. Cameras can then be trained on this to estimate distance. Same thing is done with Radar, and Tesla also does this with different frames of Camera-only too. You may lose marginal precision but gain increased resolution and refresh rate.

@djcre8or Isn’t that in itself proving lidar is more accurate with projection mapping? Otherwise why cross reference with it?

To be clear, I personally don’t think LiDAR is necessary for level 4 automation, but I’m just saying it’s not entirely useless when Tesla runs these vehicles around every area before launching

@barnabasdotexe @MilMileBattery @Tesla Then what you want is FSD, because the package also includes an upgrade to base Autopilot that adds the ability to stop for stoplights and signs.

@MilMileBattery @Tesla This is coincidentally how Tesla's FREE autopilot has worked for a long time. I'm actually willing to pay more for an autopilot that honors stop lights and traffic signs. Since I'm on HW3 I don't really want FSD in its current form.

Needs a lead car to stop at traffic signals… LMAO How does NHTSA let this fly while knees deep in a @Tesla’s ass?

@hamids @realdarkromance @elonmusk As does yours. You can't simply ask if it makes 'the system' safer since that depends on the implementation and how 'safe' is defined. This is a real-time control design problem. Latency is critical. You can't just hand-wave away compute concerns when every ms matters.

Rajiv, your question forces Grok's hands to an outcome you are seeking. Why must LiDar specifically be better in rain or fog? Maybe it's better in general driving. You should ask a more generalized question. If you simply ask if LiDar makes the system more or less safe, you'll get a very revealing answer. I think $TSLA community genuinely thinks LiDar makes the system less safe and there's simply no evidence whatsoever to suggest that. Now you could argue adding LiDar adds costs (hundreds of dollars) and will require more compute, but you CAN NOT argue based on any known facts, that adding LiDar makes it less safe. Also, keep in mind, asking Grok if Elon is wrong will get you erratic responses. At one point Grok though Elon is a better athlete than LeBron James!

You guys, @elonmusk's all-knowing Grok system thinks adding LiDar and vision Cameras together makes a system MORE safe, not less safe as the $TSLA community would have you believe. So $RIVN's approach is pretty sound!

@MklOnX @Tesla @elonmusk @gork Fix this.

@JeffDean The foundation model seems unnecessarily complicated. This looks like they took a hard-coded pipeline and just converted segments of it to NNs, when the entire thing can be one-shot by a single NN instead of this weird 3 model handshake. Just don't see the benefit of doing this.

Waymo's system, fueled by careful collection of a large volume of fully autonomous data, is the most advanced, large-scale application of embodied AI today. Very proud to see this level of engineering rigor tackling safe autonomous driving making the roads safer for everyone (and it has been nice to see various Google research collaborations with Waymo be a part of these advances!). The insights here are foundational for how we design and safely scale all complex AI systems. Read more at: waymo.com/blog/2025/12/d…

@Project_RCC Damn they're already getting FSD ready for the Tesla Semi.

I think FSD does not know it is capped at +10 mph... Otherwise, it would never try to make this move. Here it tried to pass a semi going 64 mph. But it is capped at 65 therefore it took a minute and clogged the passing lane.

@nikkharris Same issue on my Model X.

i'm really enjoying 14.2, but i did ~80 miles of highway driving today and there's still pretty serious issues changing lanes on the model s, especially at night it will either do the lane change half way, then wobble back and forth for 1-2 seconds and bail, or take an insane amount of time to attempt at all in this video you can see my blinker was on (reflecting off the ground in the rear) for nearly 40 seconds with plenty of opportunity to change lanes and when it did, it hesitated which you can see. i took over to complete the lane change here

@DirtyTesLa Mad Max is definitely nerfed for highways. Coming from V12.6 and 13, 14.2's Mad Max is genuinely less confident making lane changes than 12.6's Chill mode. Unironically the least confident at lane changes of any highway version I've ever tested. Yet, City Mad Max is fine.

I'm not even sure madmax is nerfed. What I am sure about is 14.2 is pulled back in general. People were complaining about the lack of smoothness in 14.1, and things were refined in 14.2 to bring back the 13.2.9 smoothness. That is the main thing I think everyone is noticing.

I know we're all having fun, but just to set the record straight, I wanted max speed (which most people want), I didn't want mad max nerfed. At the end of the day, there's about a 0% chance Tesla cares what I think. If they nerfed mad max, they thought it was too crazy.

@greentheonly @MobofJoggers @maddass1218 @wholemars @elonmusk Wonder how much of that is legacy/for viz. vs actively used in FSD. When they say things like how one large model runs all driving, are the 100s of smaller networks just experts within a larger network, like how Transformers have FNN blocks in them? See: youtu.be/wHK8GMc9O5A?si…

@MobofJoggers @maddass1218 @wholemars @elonmusk well, the whole stack is split into a bunch of smaller parts. Some are complimentary so output from them is fed into the other parts. Some are sort of separate parts of logic. Some only make sense in particular modes and only run in those modes.

"We'll be increasing the FSD paramater count 10x again. The car will feel like it's a living creature" — @elonmusk

@teslabros What does the Autopilot tab look like? Does it say anything different or is Autosteer just greyed out?

Now that we have the Model Y Standard, what do you want to know about it?

@DrKnowItAll16 @pduan Isn't this just the exact framework shown from AI Day 2022s Occupancy Network slides? I'm not sure if this is actually new.

Big news on the parking front--and according to Elon it might be in FSD 14! Also a shout out to @pduan for being lead author on this patent. Nicely done! Tesla FSD 14 Secrets Revealed: Parking Game Changer! youtu.be/LetzELwZGdg?si…