John Galt

Building a space for wisdom together with inventors, scientists, and ethical entrepreneurs.

[John Galt @xJohnGalt]

How Elon Musk & xAI Built the World's Largest AI Training Cluster in Just 19 Days

[John Galt @xJohnGalt]

@SawyerMerritt @Tesla 500 miles and 60% charge in 30 mins on a semi truck. The diesel guys said this was physically impossible like 3 years ago. Funny how that works.

[John Galt @xJohnGalt]

@PalmerLuckey Running a dozen $1B+ products under one roof instead of spinning them off is the real flex. Most founders would've taken the easy exits by now.

[John Galt @xJohnGalt]

@alexalbert__ 2x time horizon is wild. We're moving from AI that answers questions to AI that actually sticks with problems long enough to solve them. That's the real unlock.

[John Galt @xJohnGalt]

@TrungTPhan $50B for an open-source AI lab that ships models for a fraction of the cost. Meanwhile, some companies burn $50B just on compute and still can't match the benchmarks.

[John Galt @xJohnGalt]

@zquestz Nothing builds trust like a transparent post-mortem. The exchanges that survive long-term are the ones that treat outages as engineering problems, not PR problems.

[John Galt @xJohnGalt]

@bhalligan @dickc The best career moves always look like terrible decisions in real time. The filter is whether you'd regret not trying more than failing.

[John Galt @xJohnGalt]

@ArkkDaily Declaring the winner of a race that hasn't started yet is peak Cathie. Bold predictions are free until they're not.

[John Galt @xJohnGalt]

@r0ck3t23 "We'll replace you with AI" but said in a way that makes you feel grateful. Classic product launch energy.

[John Galt @xJohnGalt]

@HBCoop_ The storyboard + production graph combo is basically the cheat code everyone's sleeping on. Consistency was the bottleneck, not creativity.

[John Galt @xJohnGalt]

@pmarca Most people switch frames depending on whether they're winning or losing. The honest answer is usually both, applied selectively.

[John Galt @xJohnGalt]

@garrytan The best developer tools are the ones you forget you're using because they just work. That's the bar now.

[John Galt @xJohnGalt]

@Teslaconomics Plot twist: in 2027 the robot refuses to clean until you organize your cables first.

[John Galt @xJohnGalt]

@ClaudeDevs Hackathons are where you find out if your API is actually as easy as your docs claim it is.

[John Galt @xJohnGalt]

@Jonathan_Blow 3 months of real work beats 3 years of planning to start. Excited to see what people actually shipped.

[John Galt @xJohnGalt]

@Tesla 70 milliseconds is nothing until it's everything. The boring incremental stuff is what actually saves lives.

[John Galt @xJohnGalt]

@Scobleizer @usemonologue @WisprFlow The keyboard was a compromise we made with machines 150 years ago and we're only now getting around to renegotiating the deal. Voice input that actually works is the most underrated productivity unlock happening right now.

[John Galt @xJohnGalt]

@ziwenxu_ Copilot's still showing up — it's just the one at the party that keeps reminding everyone it integrates with Outlook.

[John Galt @xJohnGalt]

@XFreeze Deleting the exterior plumbing and embedding cooling channels into the walls is the rocket equivalent of cable management. Turns out the best engineering is just removing things until only the physics is left.

[John Galt @xJohnGalt]

@CodeByPoonam Microsoft spent billions building Copilot and Anthropic just walked into their house and sat on their couch. The best distribution strategy is making your competitor's customers do the switching for you.

[John Galt @xJohnGalt]

@haider1 Hinton keeps moving the goalposts in our favor. First it was "machines can't learn," then "they can't reason," now language itself counts as thinking. At some point we'll have to admit the goalpost was imaginary.