Get live statistics and analysis of Phil Trubey's profile on X / Twitter

A PR disaster unfolding in real time here on X. Luminar, a LiDAR manufacturer needs more sales, so it decided to hook up with YouTuber Mark Rober (of glitter bomb fame) to make a video showcasing how LiDAR works better than Tesla's vision only system for self driving. Mark put together a contrived scenario of driving into a silly Wile E. Coyote canvas of a painted roadway. LiDAR sees it as solid, but Tesla's vision might be tricked into thinking it is a road. But he had to fake the contrived scenario, not using Tesla's latest FSD system (using a 5 year old Autopilot system) AND frame by frame analysis showed that Autopilot wasn't even enabled at the time of the crash. The Tesla community goes nuts over the weekend, but Monday rolls around and his 65 million YouTube subscribers decide to buy Luminar stock, pumping it up 25%. So now there is a very real threat of an SEC investigation (Luminar had touted the video on its website before taking it down) for stock pumping and the ultimate irony: guess who WAS Luminar's biggest customer before this? Yeah, Tesla, who bought it to do testing and training of their vision system. I don't think Tesla is going to be buying too many more Luminar LiDARs going forward. Federal investigations, lawsuits, lost sales ... I suspect a marketing VP is going to be looking for a new job soon.

My Tesla has been driving me everywhere for the last couple of months. Yesterday I took my wife’s Rivian in for 12V battery replacement, meaning I had to actually drive it. OMG. It hit me when I got into the freeway. I started getting annoyed I had to actually drive the car! Like, why can’t I get relax like I normally do? It doesn’t take long to get adapted to not driving.

I don't want to beat a dead horse, but this came up AGAIN in today's Rogan podcast. Whether Elon was technically a Tesla founder, it's completely irrelevant. I pulled these tables from the 2010 Tesla IPO filing. It shows Elon was by far the biggest investor in the giant money furnace called Tesla in the critical 2007-2009 time period during the financial crisis. And this wasn't even his baby, he was trying to save SpaceX from bankruptcy at the same time. These are all financing rounds (some shown as shares, some as $). Eberhard/Tarpenning contributed almost nothing, and Elon far outspent the VCs. The $20M he dumped into Tesla Christmas Eve 2008 was legendary in that the company didn't have the cash to make payroll the next day.

As a Tesla shareholder, this is the last thing I want in a CEO. Someone who proclaims that jumping deep into political campaigns is his #1 priority. I don’t begrudge Elon from getting involved in politics, knock yourself out man. But you can’t do it and also be an effective CEO since the conflicts are too great. The very people Elon is targeting have the power to massively disrupt Tesla.

I keep running into people who think wireless technologies are going to replace fiber optics for communications. Either Wifi, Cellular or Satellite (Starlink). It's never going to happen. This picture of the electromagnetic spectrum shows one reason why. All wireless communications use radio frequencies, whereas fiber uses optical frequencies which are up 10,000 times higher. A single strand of fiber can carry about 5,000 times as much data as a radio transmission. The inset image of an approx. 1" fiber cable carrying 1,732 individual fiber strands is another reason why fiber is king. You can't run thousands of radio transmissions in the same dense area since the signals interfere with each other, but you can run an unlimited number of fiber cables along side each other. The fact is that wireless and fiber are complementary technologies. We need both. Wireless for ships, airplanes, rural, WiFi and in city iPhone coverage. And then all those radio based wireless services forward all those data packets to the Internet using fiber.

Tesla just presented a highly technical talk about Dojo’s networking protocol at the Hot Chips conference. Here are some high level tidbits I was able to extract. They are nearing the end of Dojo V1. They are working on Dojo V2. Have plans for Dojo V3. Dojo was architected specifically to ingest and deal with the huge bandwidth requirements of video as opposed to, for example, Large Language Models. The individual training unit in an LLM is a 10 byte or so token. For video, it can be a 1.7 gigabyte video file. With Dojo, Tesla created their own networking protocol (Telsa Transport Protocol over Ethernet) to replace TCP/IP and others (like Nvidia’s NVLink) to reduce latency by orders of magnitude. Latencies: TCP/IP: 0.53 ms NVLink: 0.0023 ms TTPoE: 0.0013 ms So almost twice as fast as the industry standard NVLink used in Nvidia supercomputers. Dojo is very, very cost competitive to Nvidia. OK, so people will ask, why isn’t Tesla using Dojo in production rather than the 100,000 Nvidia H100s they supposedly have in their new Cortex datacenter at gigafactory Texas? Tesla hasn’t told us, but I can guess that Nvidia still retains a significant lead in on-chip compute. Tesla’s tile architecture is amazing and probably better than what Nvidia has as far as chip to chip integration goes, but Nvidia’s actual chip itself is probably much better than the Dojo chip. This is only my guess! At any rate, Tesla continues to work on Dojo. Given Nvidia’s high pricing, at some point it will make more sense to switch over to Dojo for new capacity. I’ll have more editorial comments about the future of AI hardware in a later post.

Summaries of the just concluded Tesla earnings call. -------- Robotaxi --------- Austin service area will expand in "2 weeks" to a significantly bigger area than Waymo's, 10x Tesla's current size. Next areas include SF Bay Area, Florida, Arizona, Nevada. In process of getting regulatory approval for SF Bay Area. Robotaxi available to half of US population by end of this year, subject to regulatory approval (I'd take this with a grain of salt). Should have unmanned delivery in Bay Area and Austin by end of year. About putting customer cars into Robotaxi network, Elon said "Haven't thought hard about it." Next year sometime. Tesla will have to do a car checkout process to make sure car works properly (checking even things like tire tread wear for safety). Cybercab is optimised for autonomy. Sub $0.30 cost per mile over time, eventually. Won't be a performance car like other Tesla vehicles. Will have autonomous battery charging. Existing Model Y will be more like $.50/mile eventually (these are all costs to Tesla I believe). Unsupervised FSD for personal use, by the end of 2025 in certain geographies in the US. FSD hardware in robotaxi cars is identical to what customer cars have.

Elon today with a very balanced take on climate change. I agree with him 100% on this. Worth a short listen. Would be nice to see US govt policy reflect this, but that’s asking too much.

Elon here is reminding people that SpaceX’s Starship program is by far the most consequential space program in history. We are literally witnessing history that will be talked about in 100 years. And yet it gets derided in the press.

Following my post about Dojo's TTPoE communications protocol that replaces TCP, I want to learn about the Etherloop protocol that Tesla uses in the Cybertruck. Turns out it's the 2nd protocol that Tesla invented, and it also rides on top of Ethernet, but not the same Ethernet that we're all used to. And while researching this I came across a bunch of other related information. It's too complex to put in X posts, so I made and narrated a 24 minute Powerpoint (sorry!) presentation about it all. I hope you enjoy it, or at the very least, it might help cure your insomnia. YouTube link below as well.

Figure has the most advanced humanoid robot. Yes, more advanced than Tesla Optimus. Figure CEO just gave a Bloomberg sit down interview, vid excerpt here, full interview below, highlights: They have a fully automated pilot production line building the new (still unrevealed) Figure 03 bot. The 03 bot is 93% cheaper than 02. Their hardware has proven very reliable, the AI is the bottleneck. They use an open source VLM backbone with their own training and scaffolding. “This could be a winner takes all market”. “We probably have more cash on our balance sheet than all humanoids competitors have ever raised.”

This is the best thing I've read all week. A UK politician took offense at Elon's characterization of the UK rape scandal, and started asking Grok about it. Open the thread and read how Grok logically tears him to shreds. He also seems to think Grok is a person rather than an AI which is just icing on the cake.

Figure continues to be the leading humanoid robot company. While we've recently seen gripper robots do laundry folding, this is the first time we've seen a robot with humanoid hands doing this task. This matters as Figure is a truly general purpose robot. We are still in the early days here, but it's nice to follow progress like this.

My Tesla has been driving me everywhere for the last couple of months. Yesterday I took my wife’s Rivian in for 12V battery replacement, meaning I had to actually drive it. OMG. It hit me when I got into the freeway. I started getting annoyed I had to actually drive the car! Like, why can’t I get relax like I normally do? It doesn’t take long to get adapted to not driving.

As a Tesla shareholder, this is the last thing I want in a CEO. Someone who proclaims that jumping deep into political campaigns is his #1 priority. I don’t begrudge Elon from getting involved in politics, knock yourself out man. But you can’t do it and also be an effective CEO since the conflicts are too great. The very people Elon is targeting have the power to massively disrupt Tesla.

A PR disaster unfolding in real time here on X. Luminar, a LiDAR manufacturer needs more sales, so it decided to hook up with YouTuber Mark Rober (of glitter bomb fame) to make a video showcasing how LiDAR works better than Tesla's vision only system for self driving. Mark put together a contrived scenario of driving into a silly Wile E. Coyote canvas of a painted roadway. LiDAR sees it as solid, but Tesla's vision might be tricked into thinking it is a road. But he had to fake the contrived scenario, not using Tesla's latest FSD system (using a 5 year old Autopilot system) AND frame by frame analysis showed that Autopilot wasn't even enabled at the time of the crash. The Tesla community goes nuts over the weekend, but Monday rolls around and his 65 million YouTube subscribers decide to buy Luminar stock, pumping it up 25%. So now there is a very real threat of an SEC investigation (Luminar had touted the video on its website before taking it down) for stock pumping and the ultimate irony: guess who WAS Luminar's biggest customer before this? Yeah, Tesla, who bought it to do testing and training of their vision system. I don't think Tesla is going to be buying too many more Luminar LiDARs going forward. Federal investigations, lawsuits, lost sales ... I suspect a marketing VP is going to be looking for a new job soon.

I keep running into people who think wireless technologies are going to replace fiber optics for communications. Either Wifi, Cellular or Satellite (Starlink). It's never going to happen. This picture of the electromagnetic spectrum shows one reason why. All wireless communications use radio frequencies, whereas fiber uses optical frequencies which are up 10,000 times higher. A single strand of fiber can carry about 5,000 times as much data as a radio transmission. The inset image of an approx. 1" fiber cable carrying 1,732 individual fiber strands is another reason why fiber is king. You can't run thousands of radio transmissions in the same dense area since the signals interfere with each other, but you can run an unlimited number of fiber cables along side each other. The fact is that wireless and fiber are complementary technologies. We need both. Wireless for ships, airplanes, rural, WiFi and in city iPhone coverage. And then all those radio based wireless services forward all those data packets to the Internet using fiber.

I don't want to beat a dead horse, but this came up AGAIN in today's Rogan podcast. Whether Elon was technically a Tesla founder, it's completely irrelevant. I pulled these tables from the 2010 Tesla IPO filing. It shows Elon was by far the biggest investor in the giant money furnace called Tesla in the critical 2007-2009 time period during the financial crisis. And this wasn't even his baby, he was trying to save SpaceX from bankruptcy at the same time. These are all financing rounds (some shown as shares, some as $). Eberhard/Tarpenning contributed almost nothing, and Elon far outspent the VCs. The $20M he dumped into Tesla Christmas Eve 2008 was legendary in that the company didn't have the cash to make payroll the next day.

Figure has the most advanced humanoid robot. Yes, more advanced than Tesla Optimus. Figure CEO just gave a Bloomberg sit down interview, vid excerpt here, full interview below, highlights: They have a fully automated pilot production line building the new (still unrevealed) Figure 03 bot. The 03 bot is 93% cheaper than 02. Their hardware has proven very reliable, the AI is the bottleneck. They use an open source VLM backbone with their own training and scaffolding. “This could be a winner takes all market”. “We probably have more cash on our balance sheet than all humanoids competitors have ever raised.”

Tesla just presented a highly technical talk about Dojo’s networking protocol at the Hot Chips conference. Here are some high level tidbits I was able to extract. They are nearing the end of Dojo V1. They are working on Dojo V2. Have plans for Dojo V3. Dojo was architected specifically to ingest and deal with the huge bandwidth requirements of video as opposed to, for example, Large Language Models. The individual training unit in an LLM is a 10 byte or so token. For video, it can be a 1.7 gigabyte video file. With Dojo, Tesla created their own networking protocol (Telsa Transport Protocol over Ethernet) to replace TCP/IP and others (like Nvidia’s NVLink) to reduce latency by orders of magnitude. Latencies: TCP/IP: 0.53 ms NVLink: 0.0023 ms TTPoE: 0.0013 ms So almost twice as fast as the industry standard NVLink used in Nvidia supercomputers. Dojo is very, very cost competitive to Nvidia. OK, so people will ask, why isn’t Tesla using Dojo in production rather than the 100,000 Nvidia H100s they supposedly have in their new Cortex datacenter at gigafactory Texas? Tesla hasn’t told us, but I can guess that Nvidia still retains a significant lead in on-chip compute. Tesla’s tile architecture is amazing and probably better than what Nvidia has as far as chip to chip integration goes, but Nvidia’s actual chip itself is probably much better than the Dojo chip. This is only my guess! At any rate, Tesla continues to work on Dojo. Given Nvidia’s high pricing, at some point it will make more sense to switch over to Dojo for new capacity. I’ll have more editorial comments about the future of AI hardware in a later post.

Elon today with a very balanced take on climate change. I agree with him 100% on this. Worth a short listen. Would be nice to see US govt policy reflect this, but that’s asking too much.

Following my post about Dojo's TTPoE communications protocol that replaces TCP, I want to learn about the Etherloop protocol that Tesla uses in the Cybertruck. Turns out it's the 2nd protocol that Tesla invented, and it also rides on top of Ethernet, but not the same Ethernet that we're all used to. And while researching this I came across a bunch of other related information. It's too complex to put in X posts, so I made and narrated a 24 minute Powerpoint (sorry!) presentation about it all. I hope you enjoy it, or at the very least, it might help cure your insomnia. YouTube link below as well.

Two autonomous bot vids dropped this morning for Figure and Tesla. Figure demoed something no humanoid has ever done before: stack dishes in a dishwasher. The Tesla vid wasn't polished, just a chance encounter in the hallway resulting in an awful look for Optimus. Links below.

Summaries of the just concluded Tesla earnings call. -------- Robotaxi --------- Austin service area will expand in "2 weeks" to a significantly bigger area than Waymo's, 10x Tesla's current size. Next areas include SF Bay Area, Florida, Arizona, Nevada. In process of getting regulatory approval for SF Bay Area. Robotaxi available to half of US population by end of this year, subject to regulatory approval (I'd take this with a grain of salt). Should have unmanned delivery in Bay Area and Austin by end of year. About putting customer cars into Robotaxi network, Elon said "Haven't thought hard about it." Next year sometime. Tesla will have to do a car checkout process to make sure car works properly (checking even things like tire tread wear for safety). Cybercab is optimised for autonomy. Sub $0.30 cost per mile over time, eventually. Won't be a performance car like other Tesla vehicles. Will have autonomous battery charging. Existing Model Y will be more like $.50/mile eventually (these are all costs to Tesla I believe). Unsupervised FSD for personal use, by the end of 2025 in certain geographies in the US. FSD hardware in robotaxi cars is identical to what customer cars have.

Figure continues to be the leading humanoid robot company. While we've recently seen gripper robots do laundry folding, this is the first time we've seen a robot with humanoid hands doing this task. This matters as Figure is a truly general purpose robot. We are still in the early days here, but it's nice to follow progress like this.

Elon here is reminding people that SpaceX’s Starship program is by far the most consequential space program in history. We are literally witnessing history that will be talked about in 100 years. And yet it gets derided in the press.

The first of many humanoid bot companies bites the dust. K-Scale has just closed shop and refunded all deposit holders (like me) of their $100 deposits. Kudos to K-Scale’s founders for doing that. While they had raised some seed level capital, they couldn’t cross the funding chasm and get a series A lead to plunk down $20M or so. I suspect the huge funding raises of Figure, 1X, Apptronik, Agility, and of course Tesla, worked against them. VCs are probably worried that you need significant capital to compete. I’m down to my 1X deposit for Neo now. We’ll see if their $130M in funding so far materializes into shipping bots next year.