Mike Larkin

3.2K posts

Mike Larkin banner
Mike Larkin

Mike Larkin

@mlarkin2012

Low-level developer. Peakbagger. Private Pilot. Founder/CTO Ringcube (acquired by Citrix) and Deepfactor (acquired by Cisco). Building hypervisors and OSes.

California, USA Katılım Aralık 2011
2.1K Takip Edilen2K Takipçiler
Mike Larkin
Mike Larkin@mlarkin2012·
@jxmnop Can confirm. Did some kernel work today and heard from Zuck within the hour. Had to go to the gym though. Told him to come back later.
English
0
0
17
1.5K
dr. jack morris
dr. jack morris@jxmnop·
Learning to write kernels might be the highest-ROI activity for displaced SWEs: → prereq: reasonable engineering ablity → six to twelve months of study → millions of dollars, mark zuckerberg showing up at your house to hire you, etc. i wish this were an exaggeration
English
27
29
1.2K
66.5K
9mmSMG
9mmSMG@9mmsmg·
Drove by a Taco Bell, and there was actually a line at the drive-through. How are people eating this stuff? Fast food is trash in general, but Taco Bell is bottom rung for me. Wouldn't eat it if it were free. What's wrong with you people?
English
1.2K
29
1.2K
157.4K
Mike Larkin
Mike Larkin@mlarkin2012·
@straceX Why would you use a database for that in the first place?
English
0
0
7
990
Gracia
Gracia@straceX·
Interviewer: 10 million players are in a live gaming tournament. Their scores are changing constantly, multiple times a second. How do you display a 100% accurate, real-time global leaderboard without destroying your database with constant sorting?
English
14
3
32
13.1K
Mike Larkin retweetledi
Rohan Varma
Rohan Varma@rohanvarma·
Had a call with Cisco leadership today to check in on their Codex deployment. They’re leveraging AI coding and codex more aggressively than most startups and frontier tech companies I talk to. Skills for everything, agent swarms to execute bigger projects, automated code review, deep system integrations, and more 👀 They’ve quickly hit new bottlenecks that we’re jumping in to help them solve with more codex. If your public company exec team isn’t pushing the org to cut project timelines by 95%, maximally leverage coding agents, and reimagine the SDLC, you’re ngmi. Shoutout to Jeetu Patel - bullish on Cisco.
English
53
37
670
94.4K
Alex Ziskind
Alex Ziskind@digitalix·
Well, what do we have here? @Dell added this little beauty to their site.
Alex Ziskind tweet media
English
72
26
419
67.3K
Mike Larkin retweetledi
CJ
CJ@MyoTech·
@SebAaltonen Here’s a thought: develop better optimized software. Even an entry level machine like the Neo is thousands of times more powerful than the consoles and computers we use to game and work in the 90s. Software bloat and inefficiency is a much larger issue than hardware limitations
English
1
2
4
296
Framework
Framework@FrameworkPuter·
@mlarkin2012 @AIatAMD You can also run it easily on Linux, which is what we usually do in Framework. Setting up on Ubuntu or Fedora is pretty simple.
English
1
0
6
500
Framework
Framework@FrameworkPuter·
The team at @AIatAMD just published their Best Known Configs for running Openclaw fully locally, and look whose hardware they featured! amd.com/en/resources/a…
Framework tweet media
English
5
22
372
22.8K
Mike Larkin retweetledi
Nic Cruz Patane
Nic Cruz Patane@niccruzpatane·
Tesla has filed a new patent for a Optimus Charging Station. The station’s load bearing tab fully supports the weight of Optimus while charging. This lets the robot shut down some (or all) of its actuators for minimal or zero idle power draw, and also helps with wear.
Nic Cruz Patane tweet mediaNic Cruz Patane tweet media
SETI Park@seti_park

🚨 @TESLA FILED A PATENT FOR A CHARGING STATION THAT HOLDS OPTIMUS UPRIGHT WHILE ITS MOTORS ARE OFF In July 2025, Sawyer Merritt posted on X that Tesla's Optimus robot "autonomously locates the nearest charging station, self-navigates to it, only uses its rear cameras to align itself and then backs into the charger to plug itself in." The clip confirmed what the robotics community had been speculating: Tesla's humanoid robot can find its way to a charger and connect without human help. What the clip did not show is what happens next. Once docked, can the robot turn off its motors? A wheeled robot can. It just sits there. A humanoid cannot. Cut power to the joints and the robot collapses under its own weight. A PCT patent published today, March 12, 2026, describes how Tesla solves this. The priority date is September 3, 2024, six days before the Cybercab wireless charging patent that published on the same day. WO2026055103A1, "Robotic Charging System, Robotic System and Charging Station for Statically Stable Charging of the Robotic System," details a charging station with an angled structural tab that slides into a receptacle on the robot's back. In the docked position, the tab supports the robot's weight in a cantilevered orientation. Motors off. Robot standing. Charging, software updates, and sensor calibration all proceed simultaneously. ⚠️ The Problem Humanoid robots are not statically stable. Unlike wheeled or tracked platforms that stay put when powered off, a bipedal robot needs its actuators running continuously just to stand [0004]. Every joint in the legs, hips, and torso must hold position against gravity. Turn off the motors and the joints go limp. The robot falls. This creates a direct conflict with charging. Keeping the actuators active during a charging session means a portion of the incoming power is diverted to balancing, not to the battery [0047]. Charging slows down, and the motors accumulate unnecessary wear. The problem gets worse when the robot needs software updates or calibration, which may require the motors to be fully disengaged [0005]. Conventional charging docks, designed for statically stable robots like Roombas or warehouse AGVs, assume the robot can hold its own position. For a humanoid, that assumption fails. Boston Dynamics has shown Atlas swapping its own battery pack. Other companies use seated or lying-down charging poses. None of these approaches allow a humanoid to charge while standing with its motors off. The patent describes the core problem: "When power to the actuators is removed, the joints can become limp and unstable, causing the humanoid robot to collapse under its own weight" [0004]. 🔩 Core Innovation 1: The tab that holds 300 lbs upright The charging station includes a base, mounted to a wall or floor, and a structural tab that extends upward at an angle between 25 and 65 degrees [0008]. The default angle is about 45 degrees [0072]. The tab has a steel or aluminum core for load bearing (rated for 100 to 300 lbs), wrapped in a polymer overmold for wear resistance and electrical insulation [0058]. The robot has a receptacle on its upper back, positioned between the shoulder structures [0049]. During docking, the receptacle slides down over the structural tab. Once seated, the tab supports the robot's weight in a cantilevered orientation, meaning the robot leans into the station and gravity locks it in place [0064]. The angled geometry prevents the robot from slipping off unintentionally while still allowing controlled undocking [0056]. With the structural tab bearing the load, the robot can shut down some or all of its actuators [0048]. This eliminates idle power consumption during charging. The battery receives the full charging current instead of sharing it with balance control. Motor wear drops. And the robot stays upright. Electrical contacts are built into the tab and receptacle surfaces. Power contacts carry charging current. Communication contacts handle data: Ethernet, CAN bus, RS-485, or USB [0083]. The two contact types are staggered along the tab so that power connects first during docking and communication disconnects first during undocking, preventing arcing and data corruption [0069]. 🦿 Core Innovation 2: Walk backward, squat to dock The docking sequence has two phases. First, the robot executes a backward walking motion toward the charging station [0101]. The rear-facing camera detects high-contrast fiducial markers on the station's surface to guide the approach trajectory [0014], [0101]. As the robot backs in, alignment arms on the station provide coarse lateral guidance, compensating for approach errors of 5 to 100 mm [0055]. The arms funnel the robot toward the structural tab. Second, the robot performs a squatting motion [0102]. This lowers the receptacle onto the structural tab. The tab's tapered leading edge guides fine alignment during this vertical descent [0057]. The squatting distance is 10 to 40 mm [0096]. The combination of horizontal approach and vertical lowering produces a two-axis docking path that locks mechanically and electrically in a single motion [0097]. Once load sensors in the robot's legs confirm that a sufficient proportion of weight has transferred to the station, the system verifies docking is complete [0103]. The patent describes this as "a backward walking motion toward the charging station... followed by a squatting motion of the robotic system" [0101], [0102]. 🛠️ Core Innovation 3: Three jobs in one dock Most robotic charging systems do one thing: charge the battery. This patent turns the docking session into a multi-function maintenance window. While the robot is mechanically stabilized on the structural tab, the rear-facing camera images the fiducial markers on the station's surface [0088]. Because the station's geometry is known, the system can recalibrate the kinematic chain between the camera and the robot's joints. This corrects for positional drift, lens distortion, or joint misalignment that accumulated during operation [0088]. The calibration routine runs while the robot maintains partial weight on the structural tab. The actuators can execute controlled motions for recalibration while the tab keeps the robot stable [0087]. Calibration results, diagnostic error codes, and system logs upload through the communication contacts to the station for review or remote transfer [0090]. Software updates and firmware patches also flow through the same contacts [0086]. The robot receives new code while docked, installs it, and is ready to resume operation when undocking completes. One docking event. Three functions: charging, calibration, and software deployment. No separate maintenance window required. When the session ends, the robot reverses the sequence: it rises from the squat, walks forward, and resumes operation with a full battery, fresh calibration, and updated firmware [0106]-[0109]. 🎯 Why This Matters Tesla's Q4 2025 earnings update confirmed that Gen 3 Optimus production lines are being installed, with volume production planned before the end of 2026. The Fremont factory is converting Model S and Model X lines to Optimus manufacturing. The target capacity is 1 million units per year. At that scale, charging infrastructure becomes an operations problem. Every minute a robot spends balancing instead of charging is wasted energy. Every calibration session that requires a human technician is a bottleneck. Every software update that takes a robot offline is lost productivity. This patent addresses all three. The structural tab eliminates idle power drain. The autonomous docking sequence removes the need for a human to connect anything. The integrated calibration and update pipeline eliminates separate maintenance downtime. The patent also describes two interface variants. In one, the station carries the structural tab and the robot carries the receptacle. In the other, the robot carries a mating tab that inserts into a channel on the station. This flexibility suggests Tesla is designing for different deployment environments, from factory walls to freestanding stations in warehouse aisles. The claims reinforce this breadth. Three independent claims cover the system, the robot alone, and the charging station alone. If a competitor builds a compatible station for Tesla's robot, the station claim applies. If a competitor builds a robot with a similar receptacle, the robot claim applies. Tesla is patenting both sides of the interface. For a fleet of humanoid robots operating across shifts in a factory, the docking station is not an accessory. It is the infrastructure that determines how many hours per day each unit can work. This patent is Tesla's answer to that constraint 🤔 🖼️ Key Figures [FIG. 1: Full system view showing robot docked with charging station, structural tab engaged with receptacle between shoulder structures] [FIG. 9A: Side view of backward walking docking approach with mating tab variant] [FIG. 5: Docking path showing horizontal approach (20-80 mm) + vertical squatting motion (10-40 mm)] 🗂️ Bibliography Patent No.: WO2026055103A1 Title: ROBOTIC CHARGING SYSTEM, ROBOTIC SYSTEM AND CHARGING STATION FOR STATICALLY STABLE CHARGING OF THE ROBOTIC SYSTEM Assignee: Tesla, Inc. Inventors: Douglas Nordman, Brian Machado, Micah Erickson Filed Date: 2025-09-02 Published Date: 2026-03-12 Priority: US Provisional 63/690,250 (2024-09-03)

English
30
142
974
78.2K
Mike Larkin retweetledi
Tony Zhao
Tony Zhao@tonyzzhao·
We raised $165M at a $1.15B valuation to stop doing demos. 2026 is about 1) deployment and 2) research. We will start shipping Memo with our new frontier models in a few months. Our series-B is led by Coatue, with Thomas Laffont joining the board. ->🧵
English
114
101
1.5K
351.1K
Mike Larkin retweetledi
tobi lutke
tobi lutke@tobi·
OK, well. I ran /autoresearch on the the liquid codebase. 53% faster combined parse+render time, 61% fewer object allocations. This is probably somewhat overfit, but there are absolutely amazing ideas in this.
tobi lutke tweet media
English
102
171
2.8K
937.5K
Mistress Dividend
Mistress Dividend@mistressdivy·
RUMOR: MAJOR CYBERATTACK UNDERWAY TARGETING AMERICA’S CREDIT CARDS IRAN IS TRYING TO WIPE AMERICANS CC DEBT TO DESTABILIZE MAJOR BANKS
English
44
5
72
5K
Mariusz Kurman
Mariusz Kurman@mkurman88·
@evilsocket Okay, just to let you know – it's a terminal multiplexer app with AI support: React UI + Rust backend. And TUIs are React-based too, at least some of them :D
English
2
0
1
131
Simone Margaritelli
Simone Margaritelli@evilsocket·
The definitive setup: Ghostty + neovim + supermaven (free) + claude code in a split.
Simone Margaritelli tweet media
English
11
10
163
15.1K
Mike Larkin retweetledi
Dr. Clown, PhD
Dr. Clown, PhD@DrClownPhD·
Second best so far! 😆
English
120
353
5.9K
370K
Mike Larkin retweetledi
MERICA MEMED
MERICA MEMED@Mericamemed·
This is why RAM costs $900.
English
712
1.9K
40.9K
4.5M