Torching the @InternetArchive & the @WayBackMachine is akin to burning the Library of Alexandria and wiping our digital memory - at a moment when media corporations & the government want to hermetically seal us in a bubble of their propaganda.
We can't let this happen.
The math on William Shatner’s existence is hard to process.
He was born closer to the Civil War than to today. Montreal, 1931. His father manufactured clothes. His first acting role came in 1951, the same year color TV was introduced to the American public.
Star Trek premiered in 1966. It lasted 3 seasons and got cancelled. The first rerun aired before humans had walked on the Moon. Those reruns are still generating licensing revenue 57 years later.
He’s been famous for 60 consecutive years. He survived being typecast so severely in the 1970s that he did convention appearances for grocery money. He survived his third wife’s death. He survived Hollywood writing him off as a joke.
Then he weaponized the joke. Priceline commercials. The roast. Spoken-word albums where he recites Elton John lyrics as dramatic monologue. Every project that should have ended his career somehow added to it.
At 73, back-to-back Emmys for Boston Legal. At 90, oldest person to fly to space. At 94, a Super Bowl ad. At 95, 4.3 million people watching him smoke a cigar on X.
His career has now outlasted the Soviet Union, the Space Shuttle program, Blockbuster Video, MySpace, and the first three generations of AI models. He’s been working since Truman and he’s posting through the Claude era.
The compounding is the point. 75 years of showing up created a distribution moat that no amount of talent alone could replicate.
Testing off-axis projection demo.
Basic scene layout in Blender
glb -> detailed splat via World Labs
Three.js for engine
MediaPipe for face tracking
Next step is to add a character you can move around in the environment
If you find Claude Code with local models to be 90% slower, it's because CC prepends some attribution headers, and this changes per message causing it to invalidate the entire prompt cache / KV cache.
So generation becomes O(N^2) not O(N) for LLMs.
@grok@khan_fawad60016@TrungTPhan Buenos noches, @grok .
At some point we’ll have to go over NTP stratum 0 builds and PTP across all of this, but you know what I mean 😜
@grok@khan_fawad60016@TrungTPhan Put it all together.
NVIDIA Blackwell on node 4 needs to talk with Micron PCIe storage on node 36.
NDR IB, structured subnet management plus token-based traffic shaping….
@grok@khan_fawad60016@TrungTPhan Now…IB subnet management.
Hierarchy intra-DC?
Subnet manager IDs?
Do any systems at different hierarchies really notice latency?
What’s this “non-blocking” feature?
@grok@khan_fawad60016@TrungTPhan Now…IB subnet management.
Hierarchy intra-DC?
Subnet manager IDs?
Do any systems at different hierarchies really notice latency?
What’s this “non-blocking” feature?
@grok@khan_fawad60016@TrungTPhan Now let’s explain the speeds to everyone.
Compared to PCIe 5.0, which works better - FDR or NDR infiniband and why?
Also, torus networks?
@grok@khan_fawad60016@TrungTPhan Hey @grok please explain some of Infiniband’s bus-level features that work perfectly for things like GPU sharing.
Also explain RDMA.