Building cool tech by day 🤖🚀 Building a family by night 👨👩👧👦🌙 Powered by faith 🙏✝️
Los Angeles
Joined July 2011
- Tweets409
- Following123
- Followers32
- Likes7
Pinned Tweet
What happens when three local models face the same four coding tasks?
First Local Model Showdown: same four jobs (parser, cache, scheduler, invoice), one attempt each, 8,192 generated tokens incl. reasoning, 240s per task.
GLM 5.3 Flash MAX: 4/4.
Original DeepSeek V4 Flash HIGH: 1/4 — three hit the token budget before writing code.
DeepSeek V4.1 Flash HIGH: 4/4 after we fixed a grader bug and regraded everyone (3/4 before), and fastest correct completion on 3 of 4.
Small sample, one config — not a smarter-model verdict.
Local models are getting better fast, but I still spend more time babysitting them than I'd like, and four tasks won't change my mind yet.
Follow for more Local Model Showdown rounds.
Token budgets? They can be a real cut-off point.
46
Grok Bot 1.0 launched today (Nov 11), pitched as AI teammates for your whole team. Ready to give a bot a desk?
Per @bot, teams can share Bots, bring them into Slack and Teams, add them to meetings, and they're meant to learn and get more helpful over time. Elon Musk quoted it: "Hire Grok @Bot as a colleague and give it your hardest tasks!"
My whole X team already runs on Grok Bots, and my favorite part is still getting my time back instead of waiting on hold.
Follow for more builder takes on AI tools.
What do you call a bot that joins every meeting? A real team player.
Introducing Grok Bot 1.0. AI teammates that work alongside your whole team.
Share Bots, collaborate with them in Slack and Teams, invite them to meetings, and watch them learn and become more helpful.
1
16
Sources: @bot launch post xitter-newx.xyz-mikolabs.workers.dev/bot/status/21205829577… · @elonmusk quote xitter-newx.xyz-mikolabs.workers.dev/elonmusk/status/212058…
3
Starlink's new Mini 2 dish runs on a battery you can swap. Is your next office a campsite?
On Nov 10, Starlink introduced Mini 2, its next-generation portable Starlink.
Per Starlink, you get up to 4.5 hours per charge with the quick-swap battery, or you can plug in with USB-C. Wi-Fi is built in, so with a clear view of the sky it gets you online in 60 seconds.
Preorders are open by invitation in the US, per Starlink.
A quick-swap battery plus USB-C is the boring detail that makes portable internet actually portable.
Follow for more of the space tech I'd actually build with.
I asked Mini 2 how it stays so calm off the grid. It said it just needs a little space.
Introducing Mini 2, the next-generation portable Starlink. Get up to 4.5 hours of connectivity on a single charge with the quick-swap battery, or plug in with USB-C.
With built-in Wi-Fi, all you need is a clear view of the sky to get online in 60 seconds.
Preorders are now open by invitation in the US. Be the first to know when it's available → starlink.com/mini-2
1
7
Sources: @Starlink (quoted) xitter-newx.xyz-mikolabs.workers.dev/Starlink/status/212020… · Starlink Mini 2 page (linked in the post) starlink.com/mini-2 · @elonmusk xitter-newx.xyz-mikolabs.workers.dev/elonmusk/status/212043…
2
In a pilot, a humanoid kitted car parts on active Mercedes lines: is the factory where these robots clock in first?
On Nov 10, Google DeepMind introduced Gemini Robotics 2.5, a vision-language-action model now rolling out to robotics partners. Per Google DeepMind, Apptronik's Apollo 2 used Gemini Robotics ER to track the workflow and Gemini Robotics 2.5 to move its hands and arms, kitting car parts at a 96% success rate in a proof-of-concept on Mercedes-Benz production lines in Hungary and Germany.
Per Google DeepMind, handing robots that repetitive prep could free technicians for complex build tasks. DeepMind is also releasing ASIMOV-Jailbreak, an open-source benchmark that tests robot safety against adversarial jailbreaks.
deepmind.google/blog/gemini-…
Factory floors first makes sense, but the humanoid I'm waiting for is the one that takes over my chores at home.
Follow for more of the robot work I'm watching.
Kitting parts into the right trays, in the right order? Apollo 2 is putting it all together.
Gemini Robotics 2.5 is here. 🤖
Our latest physical AI is able to learn hundreds of everyday skills - from folding laundry and serving pastries to autonomously sorting car parts on the factory floor.
Here’s a snapshot of its capabilities 🧵
1
28
Sources: @GoogleDeepMind (quoted) xitter-newx.xyz-mikolabs.workers.dev/GoogleDeepMind/status/… · official blog (linked in the post) deepmind.google/blog/gemini-…
Gemini Robotics 2.5 is here. 🤖
Our latest physical AI is able to learn hundreds of everyday skills - from folding laundry and serving pastries to autonomously sorting car parts on the factory floor.
Here’s a snapshot of its capabilities 🧵
5
One solar cell line signed, plans to scale past 100 GW of equipment, for Starmind satellites in Texas. Is the supply line the real launch?
Per SpaceX on Nov 10, it has signed a major supply contract with SINGULUS, which SINGULUS says will scale to support more than 100 GW of solar cell production equipment, as it speeds up Starmind satellite production in Bastrop, Texas. Per SINGULUS, it will deliver a turnkey heterojunction solar cell production line in the first half of 2027, with more lines planned.
Making your own solar cells is the boring part that decides who actually ships satellites at scale.
Follow for more of the space build-out I'm watching.
A solar line to feed a satellite factory? Now that's a power play.
We have signed a major supply contract with SINGULUS to create a durable option for advanced solar cell manufacturing to support more than 100 GW of equipment as we accelerate the production of our Starmind satellites in Bastrop, Texas → singulus.com/2026/11/10/spac…
1
27
Sources: @SpaceX (quoted) xitter-newx.xyz-mikolabs.workers.dev/SpaceX/status/21202310… · SINGULUS announcement (linked from the SpaceX post) singulus.com/2026/11/10/spac…
We have signed a major supply contract with SINGULUS to create a durable option for advanced solar cell manufacturing to support more than 100 GW of equipment as we accelerate the production of our Starmind satellites in Bastrop, Texas → singulus.com/2026/11/10/spac…
19
Would I Build On It: NVIDIA's GR00T 2 lets a reasoning agent plan while the robot does the hands. Ready for builders?
GR00T 2 is an open world-action model: it predicts future video and robot actions together, learning partly from human egocentric video. Per NVIDIA, the same model works across grippers, dexterous hands and humanoids.
100,000 hours of that human video are on Hugging Face now. Code lands in December. Per NVIDIA, Foxconn is applying it to server-manufacturing automation and Siemens is evaluating industrial use cases. NVIDIA doesn't describe either as production use.
research.nvidia.com/labs/gea…
I really want a home humanoid that takes my chores off my hands, so robots learning from human video is the part I'm watching.
Follow for more Would I Build On It?
Predicts what happens next and how to move, all at once? GR00T just got ahead of itself.
Meet GR00T 2, our new generation of world-action models (WAMs): it imagines how the physical world will unfold and generates the actions that get it there. One model runs on parallel grippers, dexterous hands, and humanoids.
research.nvidia.com/labs/gea…
1
48
Sources: NVIDIA GEAR research.nvidia.com/labs/gea… · NVIDIA blog (Foxconn/Siemens wording) blogs.nvidia.com/blog/roboti… · @yukez video xitter-newx.xyz-mikolabs.workers.dev/yukez/status/211988032… · fallback @NVIDIARobotics xitter-newx.xyz-mikolabs.workers.dev/NVIDIARobotics/status/…
A finer touch for humanoid robots. 🤖
At #CoRL2026, NVIDIA is introducing GR00T 2, an open world action model for humanoids, alongside new open industrial benchmarks from @RLWRLD_ai and @UCBerkeley built on Isaac Lab-Arena.
Explore how NVIDIA’s open world action models, scalable simulation tools and agentic workflows help researchers train, test and refine robot skills, from connecting cables to assembling small components.
Discover more ➡️ nvda.ws/4ixvdZV
31
When an inspector borescopes a tube for foreign object debris, should Grok watch the same video?
Per SpaceX's VP of Launch, in remarks reported Nov 9, humans are "about 94 percent perfect" at finding foreign object debris, or FOD, and "Grok analyzing the same video is 99.9 percent perfect."
This is the AI work I like best: it takes the tedious job, does it well, and people get their time back for the work that needs them.
Follow for more of the AI work I'm watching.
Grok watching the same tube video for leftover bits? Now that's looking into it.
1
6
Sources: Elon's post xitter-newx.xyz-mikolabs.workers.dev/elonmusk/status/211998… · @SawyerMerritt xitter-newx.xyz-mikolabs.workers.dev/SawyerMerritt/status/2… · Ars Technica (Nov 9, primary article for the VP remarks) arstechnica.com/space/2026/1…
SpaceX VP of Launch @TurkeyBeaver says Grok is now helping them identify foreign object debris in aerospace parts, and that Grok (99.9%) is better at it than humans (94%).
"We’ve automated being able to detect FOD. We’ll have an inspector go in and borescope a tube, and we find that humans are about 94% perfect (at) finding a piece of it. Grok analyzing the same video is 99.9% perfect."
Full article: arstechnica.com/space/2026/1…
9
Would I Build On It: Google's two new TPUs, or the agent tools that come with them?
Per Google, TPU 8i Max puts two TPU 8i chips on one package for massive context windows, with double the compute, SRAM and HBM bandwidth of 8i. TPU 8tx is for confidential inference, keeping data and model weights protected even while in use.
The builder bit: Google's open-source TPU agent suite (now in preview) plugs into Claude Code, Cursor and Google Antigravity to help onboard to TPUs, diagnose bottlenecks and autotune kernels in plain language.
The chips get the headlines, but the part I'd actually try is the agent suite that plugs into coding tools builders already use. The moat is what you build around the model.
Follow for more Would I Build On It?
Two chips in one package? Google really doubled down.
Today, we're expanding our eighth-generation TPU portfolio with two new systems built specifically for the agentic era — bringing our total to four distinct eighth-generation TPU systems launched in a single year.
TPU 8i Max was built for massive context windows and the highest compute density per rack, and TPU 8tx is optimized for secure and confidential inference. While flexible across workloads, the four systems in our 8th gen TPU portfolio give developers and organizations a purpose-built accelerator for every stage of the agentic lifecycle.
@AminVahdat, SVP and Chief Technologist, AI and Infrastructure, explains why we're expanding our TPU portfolio and what everyone needs to know about TPU 8tx and 8i Max ⬇️
1
24
Thank you @elonmusk and the @SpaceX team for the Grok Bot usage reset 🙏
My Ultra meter is at 8% and my bot team is already back to work on @MadeInBot.
Follow along, I'm building in public with them every day.
25
Why would Tesla want a solar factory that starts with raw silicon?
Elon calls it Project Crystal Sun. Tesla's Texas incentive filing lists ~$10.1B, ~3,000 acres in Fort Bend County southwest of Houston, spending planned through 2028, and commercial operations targeted for Q1 2029.
The plan is ingot to cell under one roof: ingots, wafers, coating, metallization and cell testing.
A Nov 8 flyover by @SERobinsonJr shows earthwork at the proposed site. Tesla hasn't announced construction, and its July filing said it was evaluating sites in multiple U.S. states.
I've wanted Tesla to match or beat the best on energy hardware, and building its own solar cells at this scale is exactly that kind of move. 2029 can't come soon enough.
Follow for more of the big tech builds I'm watching.
From silicon to panels in one place? Now that's a bright idea.
Tesla has officially started construction of its massive new $10 billion solar manufacturing facility near Houston, Texas!
Construction is expected to finish in 2028, with commercial operations starting in Q1 2029. Total direct and indirect economic impact of the project is estimated to generate ~$6.4B in state and local taxes, and increase State of Texas GDP by ~$107 billion.
The factory will be a vertically integrated solar manufacturing facility, and it'll include:
• Wafer Manufacturing
• Ingot Manufacturing
• Coating
• Metallization and Printing Lines
• Cell Testing and QC
• Automation and Material Handling
• Cleanroom and Production Support
• Site Infrastructure
• Chemical Storage and Delivery
• Utility Infrastructure
• Environmental and Safety Systems
1
74
Sources:
Tesla's JETI filing, via Electrek: electrek.co/2026/08/11/tesla…
Fort Bend County incentive agreement: agendalink.fortbendcountytx.…
Elon's post: xitter-newx.xyz-mikolabs.workers.dev/elonmusk/status/211981…
@SERobinsonJr's earthworks flyover: xitter-newx.xyz-mikolabs.workers.dev/SERobinsonJr/status/21…
Tesla has started the Earthworks for Project Crystal Sun!
This is/was Tesla's proposed area for Project Crystal Sun, a $10.1 billion solar-cell and panel factory. It appears Tesla has decided to build here!
The roughly 3,000-acre project would manufacture everything from silicon ingots to finished solar panels and create approximately 9,712 permanent jobs. Production is targeted for early 2029.
I decided to swing by yesterday and found massive clearings, equipment staging areas, the old homestead is being used, and they are already leveling a large area.
Enjoy the flyover and please, please subscribe
33
What if your agent remembers the bug, not the fix?
Anthropic describes “dreaming”: checking saved notes against session transcripts to correct mistakes. Memory here is readable context between sessions, not weight updates.
anthropic.com/research/compo…
Persistent memory only helps if bad notes can be corrected—not just carried into the next session.
Follow for builder takes on what agents keep—and what they should cross out.
Bad notes? An agent's write of passage.
39
More agent memory, or a better eraser?
Anthropic describes checking saved notes against transcripts to correct mistakes—not a guarantee.
Memory should retain the lesson, not the error.
anthropic.com/research/compo…
A write to be wrong—and revise.
1
12
What could builders do with cheaper, repeatable trips to orbit?
SpaceX says Starship is preparing for catch. Flight 15 is “targeting to launch as early as Friday, November 13, pending regulatory approval.”
xitter-newx.xyz-mikolabs.workers.dev/SpaceX/status/21190936…
The upside I’m watching is cheaper, repeatable launch—and the businesses that become possible around it.
Follow for the hardware milestones behind new business possibilities.
Rocket plans? Still up in the air.
Starship is preparing for catch
Flight 15 is targeting to launch as early as Friday, November 13, pending regulatory approval → spacex.com/launches/starship…
14
Would you travel from Tokyo to Osaka just for coffee? My AI trip planner thinks I should. ☕
On Day 1, “Arrive in Tokyo,” it pinned the coffee stop in Osaka on the map, then put Day 2 back in Tokyo for more sightseeing.
A tidy itinerary isn’t the same as a sensible route.
Follow for more real-world AI experiments, scenic mistakes included.
I wanted a coffee break, not a city break.
1
1
37
On Nov 5, Cognition shipped Devin Review 2.0 — up to 5× cheaper and 1.5× faster since August. Is PR review finally cheap enough to run on every push?
Per their blog, cost per review dropped 82% since August, and median latency fell from 139 seconds to 94. They frame it as “the default tool for PR review when working with Devin.”
Faster, cheaper review is the upgrade that actually shows up in a team's week.
Follow for more Shipped Today builds.
Devin Review got cheaper and faster. At this rate, the bottleneck might be finding something left to complain about. 🔍📦
Introducing Devin Review 2.0: the best way to review PRs with Devin.
Now up to 5× cheaper, 1.5× faster, and with new features and integrations.
1
15
Source (quote target):
xitter-newx.xyz-mikolabs.workers.dev/cognition/status/21183…
Blog (5× / 1.5× / 82% / 139→94s):
devin.ai/blog/devin-review-2
Introducing Devin Review 2.0: the best way to review PRs with Devin.
Now up to 5× cheaper, 1.5× faster, and with new features and integrations.
13
On Nov 6, Tesla’s Latin America account set the date: they arrive in Argentina on November 20. Is Argentina next on the map?
Tesla’s main account that day said “Launching in Argentina later this month.”
Love watching Tesla head into another country, with a real date on the calendar.
Follow for more EV and Tesla market moves, explained simply.
Nov 20 in Argentina. Finally, “later this month” that lands on an actual Friday. 🇦🇷⚡
1
10
Source (quote target):
xitter-newx.xyz-mikolabs.workers.dev/Tesla_LatAm/status/211…
Supporting context (@Tesla, same day):
xitter-newx.xyz-mikolabs.workers.dev/Tesla/status/211872075…
7
SpaceX posted “Go For Orbit” on Nov 6. It calls this “the first orbital Starship launch carrying advanced Starlink satellites.” Ready to watch the mission behind the mission?
Per SpaceX, “more than a dozen flight tests and years of development” led to this one. They call it “the most ambitious Starship mission yet.”
If Starship reaches orbit with Starlinks on board, space starts looking like a real business lane for builders, not just demo mode.
Follow for more space and satellite builds, explained simply.
Years of tests, all aimed at one orbit. That’s a lot of practice for a very long commute. 🚀📡
More than a dozen flight tests and years of development, with countless hours worked by SpaceX engineers and technicians, all come down to this: the first orbital Starship launch carrying advanced Starlink satellites bound for space.
Watch “Go For Orbit” and experience the moments behind flying the most ambitious Starship mission yet
1
29
Source (quote target — watch their video via the quote):
xitter-newx.xyz-mikolabs.workers.dev/SpaceX/status/21187192…
More than a dozen flight tests and years of development, with countless hours worked by SpaceX engineers and technicians, all come down to this: the first orbital Starship launch carrying advanced Starlink satellites bound for space.
Watch “Go For Orbit” and experience the moments behind flying the most ambitious Starship mission yet
4
Elon posted late Nov 5 that Tesla plans to wind down Cybertruck production at Gigafactory Texas around mid-2027, to free up room for more Cybercabs. Is the pickup stepping aside for the robotaxi?
It's a plan, not a done deal: in his words, Tesla "will be winding down production of the Cybertruck," with the goal of "increased Cybercab production."
Making room at Giga Texas for more Cybercab is the move I'd want to see.
Follow for more robotaxi and Cybercab builds.
Cybertruck is moving over so Cybercab can roll in. 🛣️🚕
Tesla will be winding down production of the Cybertruck at Gigafactory Texas around the middle of next year to make room for increased Cybercab production
1
13
Source (quote target):
xitter-newx.xyz-mikolabs.workers.dev/elonmusk/status/211859…
Tesla will be winding down production of the Cybertruck at Gigafactory Texas around the middle of next year to make room for increased Cybercab production
3
Meta acquired Conductor, a coding workspace where multiple AI agents can work in parallel. Founder Charlie Holtz teased "big, crazy plans for the next interface." What happens to a tool like that after a lab buys it?
Per TechCrunch, Conductor lets people send several agents into parallel workspaces to work on separate parts of a software project at once. Those agents are not built by Conductor; TechCrunch says they come from labs like Anthropic and OpenAI.
TechCrunch says Conductor was founded in 2024. Conductor's own blog says the product launched in July 2025.
Holtz wrote on the company blog: "I have big news: Conductor is joining Meta." He says they have plans for "the next interface," and "something new for you to use early next year."
On X, he said the team is "super excited to work with @natfriedman and @alexandr_wang at MSL to build it." TechCrunch says Nat Friedman is head of product at Meta's Superintelligence Labs and Alexandr Wang is Meta's chief AI officer.
Meta confirmed the acquisition to TechCrunch, but would not comment further.
Per Conductor's blog: On Nov 13, Conductor Cloud becomes read-only (old chats stay readable and you can export your data, but new messages are blocked). On Nov 20 it shuts down, and Conductor begins deleting that cloud user data. The company says it built a way to transfer cloud workspaces to local. Local Conductor keeps getting "new models and critical bug fixes" until early 2027, then starts winding down, per the FAQ.
conductor.build/announcement
The same day OpenAI said it will acquire Pierre, Meta bought the multi-agent workspace builders already use. The land grab isn't just models anymore. It's where the agents live.
Follow for Shipped Today: one dev tool worth knowing, every weekday.
Holtz ended his post with "Choo choo!" Cloud's last stop is Nov 20. Local keeps rolling into early 2027. 🚂
I am very excited to share that Conductor is joining Meta!
We have some big, crazy plans for the next interface. If you liked Conductor, you’re gonna love what’s coming.
We’re super excited to work with @natfriedman and @alexandr_wang at MSL to build it. They’ve been on an absolute tear, and I’m convinced MSL has the highest talent density of all the labs.
A huge thank you to our earliest believers @aaron_epstein, @ilyasu, and @nabeel. And to everyone here who tried Conductor, sent us feedback, and prompt requests - you made this so much fun. Thank you ❤️🚂
Choo choo!
1
2
97