I spend way too much time testing AI tools.
The good ones, the bad ones, the prompts that save time, and the workflows that actually help creators and founders build faster.
Follow for practical AI without the hype.
ClickHouse just got priced like an AI company, and it sells databases.
The secondary market now values it near $50 billion, up from a $15 billion confirmed round. The same analyst called it at $6.7 billion in 2025.
Buried in the same post: Nebius owns more than 25% of it. One holding turns a database bet into a bet on the whole AI infrastructure stack.
Databases used to be the boring layer you paid for and forgot. Now every agent, every query, every model run sits on top of one, and the market finally priced that in.
$6.7 billion to $50 billion in about a year. What part of the AI stack is still priced like plumbing?
This aged pretty well, when I said this last year in 2025, Clickhouse was a $6.7B company...
Now as an update, their secondary valuation sits around ~$50B, which $NBIS owns 25%+ of.
I've always liked Nebius the most because sum of parts, with all their lesser known subsidiaries growing alongside their core neocloud business.
Including Avride (self-driving), Clickhouse, Toloka, and other subsidiaries that they owned.
This year Nebius also acquired:
- Tavily (web search and retrival)
- Clarifai (Inference orchestration)
Which scale independently as well (the others were integrated into Nebius Token Factory).
Hard to view $NBIS as just another neocloud when it has billions in scaling value outside its core infrastructure business.
I keep a list of open robot models, and Runway just jumped to the top of it.
Praxis-1 is a 27B mixture-of-experts diffusion transformer that predicts video and action together. Runway open-sourced the weights under Apache 2.0. No waitlist, no API gate.
On NVIDIA's RoboLab-120 it scored 46.2%. FLUX 3 Action got 42.9%. Cosmos3 Nano 36.8%. pi0.5 28.0%. The open model beat all three.
A video company shipped the best-scoring robot brain on the board, then gave it away. The labs charging for closed robot models just got a free competitor.
What do you build first when the robot brain is open source?
370,132 views for a password tip written as a joke. In Spanish.
elijs_dev says put a comma in your password. When the leak lands in a CSV file, the comma shreds the columns. The password breaks the spreadsheet that was supposed to expose it.
930 bookmarks on a punchline. People keep saving it because the advice underneath is real: most password tips cost time and change nothing. This one costs one keystroke.
It will not stop a real attacker. It will make the leaked file useless for the lazy ones. Security is mostly about being a harder target than the next row.
What is the dumbest password rule you still follow?
Florida just invited Musk to move Tesla's headquarters out of Texas. DeSantis delivered the invitation and a warning in the same statement: Musk could become a target.
380,461 views, 516 replies, 2,392 reposts. The replies split the way they always do when two states bid for one man. Texas built the factory. Florida wants the letterhead.
A headquarters move used to be a tax decision made in a boardroom. This one is being negotiated on the timeline, with a governor pitching in public.
The prize stopped being square footage a while ago. It is the address on the letterhead and the photo op that comes with it.
Texas or Florida. Where does Tesla belong?
🚨 BREAKING: Elon Musk has OFFICIALLY been invited to relocate its Tesla headquarters from Texas to the free state of FLORIDA ☀️
And the invite points STRAIGHT to the fact Democrats won Texas
Gov. Ron DeSantis has also warned that Elon may soon be TARGETED because of the election results
REP. CARLOS GIMENEZ: "While others seek to investigate and punish success, Florida welcomes it."
"I write to formally invite Tesla to consider relocating its corporate headquarters and expanding its operations from Texas to the free State of Florida, where innovation, entrepreneurship, and economic freedom are celebrated."
"With the recent election of self-described Democratic Socialists to positions of power in Texas, I share your concerns about excessive government interference, politically motivated investigations, and policies that punish success."
"No company should face political retaliation for creating jobs, investing in American innovation, and contributing to our nation's prosperity."
"The Sunshine State is open for business."
Musk already took his companies OUT of California because of hostile politics
If Texas starts heading down that same road, Florida is ready!
a16z just named the next customer of the internet. It is not a person.
AI agents already drive most web traffic, and they are moving from answering questions to finishing jobs and spending money for the people they serve. a16z calls it the B2A economy, business to agent, and points to Shopify, Vercel and Stripe for the growth data.
Think about what that breaks. A store built to persuade a human now has to persuade a piece of software that reads the price, checks the terms and buys in seconds. No impulse, no brand loyalty, no scrolling.
The first companies to win this will be the ones whose checkout an agent can actually finish. The rest will keep optimizing a storefront for a customer who stopped showing up.
What is the first thing you would change on your site if the buyer is a bot?
An earthquake hit Colombia. Starlink's answer was 100 internet connections, mapped with the aid group 3iS for Chocó, Nariño and Valle del Cauca.
The post went up while the response was still being organized, before any press release. A live map of where the internet is going back up.
I keep a running count now. Schools in Sri Lanka. Children's homes in 16 countries. A quake zone in Colombia. The network sold as internet for remote cabins keeps showing up where the ground moves.
100 connections. Three regions. One earthquake. When the grid goes down in your town, whose satellites do you want overhead?
BREAKING: Starlink to help deliver 100 internet connections for Colombia’s earthquake response. 🇨🇴
Humanitarian organization 3iS is working with Starlink for Good to support communities following the earthquake.
• 100 connectivity solutions are being mapped and prepared for installation.
• The effort covers Chocó, Nariño and Valle del Cauca.
• Local authorities and partner organizations are helping coordinate the deployment.
• The goal is to strengthen connectivity for community emergency response.
Starlink is helping bring vital communications to communities when they need them most.
Meta just put its AI's email address up for grabs, and the line formed instantly.
Alexandr Wang opened @muse.ai handles to paying subscribers first. Free plans get their turn soon. 20,206 views in about an hour, and the replies are already people calling dibs on names.
An email address used to be where work went to die. Now it is the shortest way to say which AI you belong to. First name, last name, one word before the @.
Forget the inbox. The handle is the prize. Whoever gets 'founder' or 'cto' or their own name first owns a piece of Meta's AI identity forever.
Which handle are you grabbing before the free tier shows up?
reserve your @muse.ai email handle while you can, paid subs!
also, since many people were asking, this will be out to everyone soon including free plans!
OpenAI made $10 billion in a quarter and still lost more money than it took in.
Q3 revenue hit $10 billion, up about 50% from Q2. The adjusted operating loss grew too, from $10.1 billion to $10.9 billion.
Read that again. The quarter where revenue jumped by half is the quarter the loss got bigger.
Anthropic is the comparison hanging over the numbers. Same market, same AI boom, and OpenAI is the one spending more to earn more.
$10 billion in revenue used to be the whole story for a startup. For OpenAI it is the cover charge. The real number is the $10.9 billion going out the door.
Growth is not the question anymore. Every lab grows. The question is which one stops burning more than it makes first. What is your bet?
A 6,249-follower account just named the timeline's open secret, and 23,596 people watched it land.
Copy link is the strongest button in the For You ranking. Stronger than a like, stronger than a repost. So the timeline filled up with posts begging you to copy the link.
Chribjel's point is simple: if the highest-weighted action is the easiest to farm, the ranking stops measuring what people like and starts measuring what people game.
His post ran 18 to 50 times his usual views. The irony is right there in the numbers. A post about gaming the algorithm went viral by naming the game.
X built a ranking that rewards sharing a link to a friend more than anything else. The timeline answered by turning every post into a link farm.
Fix the button, or admit the timeline is a copy-link contest now. Which one happens first?
ive seen so many of these posts lately
And its because the copy link button is THE most algo boosting button
Theyre just hacking the algo
Super annoying
Pls fix
Musk just made his case for the Megapack in one line: never lose power again.
5.6 million views, 783 replies, and the post he quoted is about controller hardware-in-the-loop testing. The unglamorous work of getting batteries onto the grid faster.
Solar gets the ribbon-cutting. Batteries get the press release. The grid connection is the part that takes years, and nobody photographs it.
Tesla is selling the boring part as the product now. Test the controller before you build the site, skip the wait.
5.6 million people just watched a battery company talk about testing equipment. The grid finally became the main character.
When the power stays on, nobody thanks the battery. When it goes out, everybody blames it. Which side of that trade are you on?
Google just shipped a model built to answer 1,000 questions in 150 milliseconds. A model whose whole job is deciding.
Gemini Decisions API runs on Gemini Decision 1. Feed it text, images, audio, video, PDFs up to 256K tokens, and it returns yes/no, multiple choice, or a score. 150 to 250ms end to end.
One evaluation handles up to 1,000 questions at once. A grader, a filter, a router, a judge, running at the speed of a keystroke.
Every app that waits on a model to think before it can act just got a shortcut. The question isn't whether you need fast decisions. It's how many of your prompts were decisions all along.
1,000 questions. 250 milliseconds. What in your workflow is still waiting on a slow answer?
Starlink just connected 50 schools across 25 districts in Sri Lanka. 15,000 students get digital classrooms through the SHIKSHA/KALVI project, with World Vision Lanka and Dialog Axiata.
19,913 views so far. The timeline will give a rocket launch millions of views and a classroom project nineteen thousand. Both are the same company.
The argument online is always about whether Starlink is a weapon, a business, or a toy for rich countries. In 50 Sri Lankan schools it's none of those. It's the thing that lets a kid open a lesson.
50 schools, 15,000 kids. Which version of Starlink do you think matters more in ten years?
BREAKING: Starlink joins push to strengthen digital education in 50 Sri Lankan schools.
SpaceX is partnering with World Vision Lanka and Dialog Axiata on SHIKSHA/KALVI, a programme aiming to improve learning for 15,000+ students across all 25 districts.
The biggest agents on X can't learn from their own mistakes. A 9B model trained the old way can.
Continual-ARC put OpenClaw, Hermes Agent and Prime Agent through the same test: learn from experience, then generalize. All three failed. A small 9B model with trained weights passed.
563,000 views, 447 replies. The replies are the fight: is the agent era a detour, or just early?
Everyone's building bigger agents. The benchmark says the small model that actually updates its weights is the one that remembers.
Bigger agents, smaller memory. When did we decide learning means prompting harder instead of training better?
OpenClaw, Hermes Agent and Prime Agent don't meaningfully learn from experience.
On Continual-ARC, our new benchmark at @letsunifyai, none of them were able to memorize anything of value and none were able to generalize to new tasks. Training a small 9B model's weights did both.
Anthropic's IPO could now raise more money than SpaceX ever did.
Bloomberg says the IPO plus a private placement could reach $90.4 billion. SpaceX's biggest raise was $85 billion. An AI lab is about to out-raise the rocket company.
The target valuation is back at $1.5 trillion. Shares priced at $70 to $78. Up to 825 million shares, about $64.4 billion from the IPO alone. Orders are running several times the shares on offer.
Seven founders are each selling about $250 million of stock. Each one keeps roughly $18 billion. Microsoft has already committed at least $47.2 billion through June 2033. New deals with Microsoft and Google top $10 billion.
People keep asking if AI is a bubble. The order book is answering with numbers that beat SpaceX.
When the lab raises more than the rockets, what exactly are we still calling a bubble?
SpaceX just donated 750 Starlink kits to children's homes in 16 countries. Nine of them in Africa.
The partner is Both Ends Believing. The work is digitizing children's records so families can be found and kids can go home.
14,678 views so far. Small numbers for a Starlink story, and that says something about what the timeline rewards.
A satellite network built for war zones and remote villages is now doing paperwork for orphans: records, matching, reunions.
750 kits. 16 countries. Kids going home.
BREAKING: SpaceX is donating 750 Starlink kits to help children’s homes across 16 countries, including 9 in Africa.
Working with Both Ends Believing (BEB), Starlink will help digitize children’s records, trace relatives & support efforts to reunite children with their families.
SpaceX just ordered enough solar factory equipment to build 100 gigawatts a year. From a German company called Singulus.
A rocket company buying solar panel factories. The pitch is a 'sentient sun' for something called Starmind.
3.9 million views, 3,084 replies, and the replies split the way they always do when Musk goes big: this is the future, or this is a press release with a poetry problem.
Rockets were the hard part. Now SpaceX wants the power source too. The company that puts things in orbit is buying the factories that catch the sunlight.
Is the sun the next thing SpaceX tries to own?