On my Mac, there’s a command last reboot, which shows that I’ve rebooted my computer only four times over the past year.
Tag: Tech
Exploring the Technological Marvels of Tesla’s Full Self-Driving Capabilities | July 11 2025, 03:59
I read various engineering blogs about Tesla’s autopilot (FSD) — simply because for the last month and a half I’ve been almost constantly riding as if in a taxi — you set the destination and hardly ever need to intervene, the car travels from point A to point B completely independently. This is certainly the future.
Such systems exist not only at Tesla. For example, Mercedes has one (Drive Pilot). Others only help in traffic jams at best. Though Tesla seems to be the only one that works on all roads.
So, returning to engineering curiosities. Tesla has an AI model production on its “farm” called Dojo — an exaFLOP supercomputer on Tesla chips. Videos from cameras are fed into it, and it trains models that are then sent out for autonomous operation across the entire fleet of Tesla cars.
The FSD architecture comprises about 48 specialized neural networks, trained on Dojo, which together form about 1,000 different prediction tensors. Tesla is gradually moving from modular networks (object recognition + planning) to end-to-end training — directly converting video frames into steering trajectory/action. This is akin to a “black box” — the neural network learns directly from human behavior, without manual tuning of knobs; an extremely cool engineering solution, but, I suspect, complex to debug.
By the way, it is claimed that Tesla has switched from C++ to Python. And that this shift to end-to-end training has made 300,000 lines of C++ code unnecessary, where various corner cases and rules for resolving different scenarios were accounted for — now it’s at the model level.
Tesla has abandoned radar and ultrasonics, switching to purely camera solutions (Vision Only) with “Hardware 4” (HW4, FSD Computer 2): 16 GB RAM, 256 GB flash memory, performance 3–8× higher than HW3.
Assess the performance: 22 milliseconds to create a 3D scene with cars, pedestrians, cyclists around — information is collected from 8 cameras 36 times per second.
85 ms for the entire cycle from receiving the image to changing the plan and commands to the wheels. Fantastic!
More than 4 million Teslas on the roads collect data daily, and in the FSD Beta version, more than a billion miles of autonomous driving have been recorded. This “live” dataset is used to train networks on the most real-world scenarios, including rare “edge-case” incidents (strange accidents, road conditions, etc.).
In June 2025, Tesla for the first time delivered a Model Y from the factory in Austin to a customer’s home without a driver or remote operator — fully autonomously. This is very cool.
The Vision network not only analyzes the current frame but also stores features from previous ones (at a distance of ≈1 m). This allows it to remember recently crossed markings/signs, even if they have already left the field of view – very similar to human memory.
Awaiting the Next Big Thing from Boston Dynamics | July 10 2025, 20:09
I’ve been thinking, it’s been a while since there was an exciting video from Boston Dynamics. Remember, each one used to create quite a stir online. I checked their channel. And it looks good. Hundreds of millions in investments from the Pentagon/US Department of Defense suggest that it won’t stop at drones.
Exploring the Bubble Method of River Level Measurement at the Potomac | July 06 2025, 19:38
How would you measure the water level in a river? A float? A pressure sensor? Something else? Yesterday, I discovered how it’s done here on the Potomac, and it turned out to be not at all what I had imagined. The USGS engineers are great—they educate passersby by posting a diagram of the operation.
A tube is lowered into the river through which air is supplied in bubbles (through a bubble orifice). A special pressure sensor (Pressure Transducer) measures the air pressure in the tube that is necessary to release the bubbles from it. The higher the water level in the river, the more pressure is required to push the air into the water—because the air pressure in the tube is directly related to the depth of the water (according to Pascal’s law). The bubble method works well even if there is floating debris or ice in the river, which may interfere with other sensors (such as ultrasonic ones). Since the sensor does not contact the water, it always remains dry and clean. Additionally, to prevent data distortion, the system includes an air dryer (Air Dryer), which removes moisture from the air and prevents condensation.
The accuracy of such systems is 1-2 cm in water level for rivers with shallow depths.
Interestingly, the readings are transmitted not through the mobile network, but via satellite.
Lost in Translation: A Midnight Encounter at Ashburn Station | July 06 2025, 17:28
Yesterday late, around 10-11pm, I was returning from Washington by metro. At the exit of Ashburn station, a relatively well-dressed guy approaches me and asks how to get to Route 7 from the station by bike. I start to answer, then he asks me if I happen to speak Russian. My accent gave me away (damn, how did he know exactly?)
I open the map on my phone, start explaining it to him, go right here, then left, then right, a 45-minute ride. It’s night outside. The dude’s on a bike. He doesn’t have a phone — something is broken or dead. But the most interesting thing, he doesn’t know the address where he needs to go. And Route 7, by the way, is 497 km long, but he obviously meant a segment about 30 km near the metro, but it was still not clear where he needed to go in that section.
In the conversation, it turned out that he knows how to get to the place where he stopped (friends?), from the local Russian-speaking Protestant church, called New Life. I feel I’m explaining to him, he’s overall ready to go alone in the dark without navigation, but from his feedback, I understand he didn’t get it, and at the first turn, he’d go wrong. And at that time, there was absolutely no one on the streets, it’s a neighborhood and data center area (the largest in the world, by the way), very safe, but absolutely deserted. I tell him — my car is parked at the metro, let me give you a lift if that’s the case, it’s no trouble for me.
His name is Edik. He wrecked his car a week ago because he liked to drive “with a breeze”. He regrets it because now he doesn’t understand what to buy a new one with. Lives in Baltimore, came to our area because there’s some Mongolian holiday tomorrow. What? I ask, what the hell is a Mongolian holiday. Turns out he’s from Mongolia, lived there before moving to the USA. Russian family, school at the Russian embassy. Speaks Russian without an accent, and fluent in Mongolian. Illegal. Apparently, he came to the USA on a tourist visa and stayed. Works in a store somewhere near Baltimore. Deep in debt. Apparently, a few adventures weren’t enough and he went to Virginia by bike mixed with metro and buses.
I hope he made it home from the church.
Advancing Full-Text Search: Testing and Refining with Multi-User Platforms | July 06 2025, 04:35
I have developed expertise in full-text search testing. Essentially, it’s a turnkey multi-user platform that, given roughly 1000 queries and several search engine configurations, can produce reports with graphs, metrics, and conclusions by morning, showing why configuration A performs better than B, and here’s why. It calculates all those NDCG@k, MAP, precision, recall, and about a dozen other metrics. It uses LLM, but only at the final stage, after all the math is done.
So, here’s my question. I’m looking for someone who has faced the same issue in their project, to understand the demand and the ask.
The problem the system solves is defined as follows: there is a functional search for goods, documents – Solr, Coveo, Elasticsearch, Algolia – it doesn’t matter, and there are hypotheses on how to improve it, but there is also the fear that improving one aspect might break another. Well, my thing helps to see this in numbers and graphs, providing a conclusion with justification, including statistical significance and other metrics.
It also acts as a virtual search assessor. For each search result, it can give a rating, assessing how well each document matches the query. This is a very non-trivial task (especially for large documents), involving chunking, embeddings, LLM evaluation of relevant chunks, etc. Non-trivial, but it works.
It also can analyze search queries and break them into groups based on similarity. For instance, such segmentation might show that users sometimes separate the words forming a brand name with a space, and sometimes not. These different variants would be grouped together.
I would like to discuss this with someone who knows more about this topic than I do, someone who has/had such problems and has somehow solved them.
I currently feel like my product is unique in the market. Actually, it’s not even on the market yet. But I really don’t see anything similar out there. Maybe nobody needs it?
I won’t publically post screenshots yet. The picture is merely for attracting attention.
Please share if there might be relevant people in your network.
A Costly Trip to the National Cryptologic Museum: Enigmas and Espionage | July 02 2025, 14:56
I went to the National Cryptologic Museum yesterday. Indeed, this trip will cost me $1000 because a rock hit the windshield of the new Tesla on the way. Anyway, let’s talk about the museum.
It’s very small. Located on the premises of the National Security Agency. The museum basically consists of three small rooms. One is dedicated to German Enigmas and there exhibits Alan Turing’s Bombe decryption machine, — a device as big as a kitchen in Lobnya, used for systematic decryption of messages encrypted by the Germans using “Enigma.” After the war, Churchill, for reasons of secrecy, ordered all physical traces of the program, including the Bombe machines, to be destroyed, so it’s quite a rare thing. Moreover, there’s only one working Bombe machine in the entire world, somewhere in England, and even that was barely restored. The Enigmas themselves were produced in large numbers, and the museum has two working ones; you can press the buttons and encrypt something.
In the room with computer equipment stands an old Cray, as well as a decommissioned nuclear deterrence hardware server rack taken out of service 15 years ago. It’s not very clear what’s remarkable about this – well yes, old computers, that’s all. The Cray is actually exhibited many places.
Unfortunately, there are no longer exhibits from the Star Gate project — like the blue box shown in the attached photos. The Star Gate project was used by the US government during the Cold War. Many of the psychic spies were based at Fort Meade, tasked with gathering intelligence, detecting enemy agents, and identifying vulnerabilities in the US using “remote viewing.”
Never heard of “remote viewing”? It’s the mental observation of a distant place where a person has never been, in order to gather information about an individual, an object, or specific data. As absurd as it may sound, it’s claimed that the program was quite successful and used until 1995 🙂
Specifically, this little blue machine, PSIFI, is part of that program. For example, it was used to study the impact of consciousness on random processes — like altering the behavior of random number generators through thought, collecting statistics on attempts at psychokinesis — with “hits”, “trials”, “gated hits”, “gated trials” etc., suggesting successful impacts compared to an expected random distribution, biofeedback — the lower part of the panel contains controls and inputs, apparently for electromyography and other biosignals. Overall, a good addition to the UFO research program.
Exploring Xplor Park: An Engineer’s Marvel in Riviera Maya | June 29 2025, 05:41
I returned from Mexico — visited Xplor Park by Xcaret in Riviera Maya. The park is already 18 years old, but damn, it’s an engineering feat, not just a park. As an engineer, I was walking around with my mouth open.
The park is the size of Moscow’s “Neskuchny Garden”. A significant part consists of kilometers of natural karst caves, formed millions of years ago at the site of the Chicxulub impact crater (the very one that ended the era of dinosaurs). Above the caves are dense jungles. High above the jungles — kilometers of zip lines. The water in the caves is from a natural underground stream, which is filtered through limestone plus some technical structures. Bats fly around, but obviously, they are not wild and are working for food. No wildlife (other than tourists and bats) is present, so it’s pretty well isolated from the outside world. In these kilometer-long caves, completely covered with stalactites and stalagmites, we swam, rafted, and even drove through in amphibious vehicles with gasoline engines (meaning, the ventilation is well-thought-out).
In front of us, three Mexican women failed to control their vehicle and crashed into a tree. Literally — the front wheels of the buggy were above my head. We picked them up walking along the track, sat them back, and about 5-10 minutes down the road, park workers took them away. The girls have something to remember.
The ticket includes a very, very good buffet restaurant. But pictures are essentially a must-buy — a very thoughtful system designed to extract about 100 dollars from a visiting family. Helmets are embedded with a chip, the system classifies the pics on the fly, and at the exit, you can see all your photos and buy them right there. And on the way back to the hotel, you can post on Facebook or Instagram.
Well, we’re back home now, back to work from Monday.
Persistent Notifications: The AirPods Pro Annoyance on a Flight | June 26 2025, 12:45
This weird thing appears on the phone and you can’t close it, it just keeps popping up again and again, every second. For about five minutes. It’s almost impossible to use the phone. Turns out, there’s a guy sitting one seat away from me on the plane, opening and closing an AirPods case, chatting with a girl. He’s got nothing better to do with his hands, darn it.
The Surprising Origins of Chain Link Fencing | June 26 2025, 10:08
Deception is everywhere. I googled “chain link fence” and it turns out that Karl Rabitz has nothing to do with it, but instead relates to a different one, and the very first of the known documented images of the chain link fence was found in… a mattress patent. More precisely, in the US patent No. 124,286 “Wire Fabrics”, issued on March 5, 1872, to a certain Mr. Peters (J. W. C. Peters).
Hi, I'm Rauf Aliev. 