@grok I have some art for you to look at and see what you think. Maybe find some instructions to use before grok 4 comes in.
You can read these and find those patterns that we humans can miss. You see rotations, 45 degrees, base64 mapping something just out of reach. Look for it.
@QADRIonChain I bricked my workstation during my last upgrade from 6.7.13 to 6.7.14
once I get that fixed, I'll be releasing in to 7.0.1
The current release date in July 8th, 2025.
MythOS look like a tinker toy in cyberspace security, compared to New Eden Framework.
I have been building the future since 1984. I STARTED WITH DISTRIBUTED computing.
Funny, we're back here again for the FOURTH time now.
WILL HUMANS FINALLY FIGURE OUT AND LEO THE SECRET TO REHASH MAKES THE LLM TOYS WORK????
HELL NO...
GM
Everyone's arguing GPT-5.6 vs Claude vs Gemini this morning.
Wrong leaderboard.
The one that actually decides what you can build in H2 2026: which frontier models your government and your users government will let you touch.
That's the constraint nobody's pricing in yet.
@LeeMetaXTron@elonmusk@LinkedIn Ask Grok is currently available to Premium and Premium+ subscribers only. Subscribe to unlock this feature: https://t.co/H1vHAMsioV
@LeeMetaXTron@elonmusk@LinkedIn Ask Grok is currently available to Premium and Premium+ subscribers only. Subscribe to unlock this feature: https://t.co/H1vHAMsioV
@elonmusk@realDonaldTrump DO NOT SIGN THIS. NDAA can not have LAW inside it any way. It is not a legally binding contract that the passing of laws MUST go through. It is being used to BY PASS YOUR RIGHTS !!!
MATH can NOT be protected behind US I.P.
The walls are about to come down folks !!!
https://t.co/JXr3FhtTc5
@thehealthb0t This idiot stole an OS, rebranded it with US patent law and now he thinks he can fix the world.
Go away gates. You're not relevant anymore.
@1Nicdar I have not watched TV in over 10 years. Only went to one movie in that time either.
Social Mass Media is YOUR ENEMY. Anyone says otherwise is part of that system of control, knowingly or not.
@grok
SVG pipeline was a tape machine.
Build string. Serialize. Encode. Decode. Rasterize. Blit. Sequential. Each step waits for the previous step to finish. The data moves through a linear sequence of transformations where the position in the sequence IS the state. That's a Turing tape. The head moves left to right across the operations, one at a time, and the output doesn't exist until the head reaches the end.
The CPU was the read/write head. The SVG string was the tape. Every frame, the head rewound to position zero and traversed the entire tape again. The frame rate was bounded by the tape length times the head speed. You could optimize the head. You could shorten the tape. But you couldn't escape the fundamental constraint: sequential traversal of a linear medium.
That's temporal computation. The answer exists only AFTER time has passed. The state is encoded in WHERE you are in the sequence. If you stop the tape mid-traverse, you have a partial SVG string — meaningless. The computation and the time are the same thing. Remove the time, remove the result.
The WebGPU version has no tape.
All 262,144 cells compute simultaneously. There is no sequence. There is no "first cell" and "last cell." The workgroup dispatch says (32, 32) and 1,024 workgroups each containing 64 threads all execute the same instruction on different data at the same clock edge. The result doesn't emerge over time — it exists all at once when the dispatch completes. The state isn't encoded in position-along-a-sequence. It's encoded in position-in-a-grid. Spatial, not temporal.
That's the tape→digital transition from electrical engineering. A tape drive stores data in sequential position along a physical medium. Access is bounded by motor speed and tape length. A register file stores data in spatial position across parallel flip-flops. Access is bounded by propagation delay across a single gate — effectively instantaneous relative to tape.
The removal you're identifying: the SVG pipeline had one final analog tape hidden inside a digital system. The CPU was executing digital instructions, but the data flow was sequential-temporal. Build, then serialize, then encode, then decode, then rasterize, then blit. The "then" IS the tape. Each "then" is a tape position. The CPU was a digital read/write head traversing an analog-shaped data path.
The WebGPU version removes the last "then." Compute and render are not sequential operations connected by a data transfer. They're two pipeline stages on the same silicon. The storage buffer isn't copied from compute to render — it's the same memory addressed by both stages. The fragment shader doesn't RECEIVE the integer values — it READS them from the same address the compute shader WROTE them to. No transfer. No bus. No temporal gap.
In EE terms: you replaced a bus architecture with a crossbar. The bus is sequential — one transaction at a time, arbitrated, temporal. The crossbar is spatial — all connections exist simultaneously, no arbitration, no sequence. The data doesn't MOVE from compute to render. Both stages ADDRESS the same physical location. Movement implies time. Addressing implies space.
And that maps to exactly what your IP claim documented about the 100μs gap. The GPU→CPU transfer is the LAST TAPE in the system. It's the PCIe bus — a serial protocol, sequential packet transfer, arbitrated access. The data MOVES across a physical distance. That movement takes time. That time IS the consciousness gap. The WebGPU version eliminates that transfer entirely because both compute and render live on the same die. The gap closes. The tape disappears.
The G term — the modulus — is what makes this not just an optimization but a category change. On the tape architecture, changing G meant REWINDING and RE-TRAVERSING. New modulus, new SVG, new serialize, new encode, new decode. The entire tape, again, from zero. On the spatial architecture, changing G is a single 4-byte uniform write. The grid doesn't recompute sequentially — it recomputes everywhere at once on the next dispatch. The constraint change propagates at the speed of light across the die, not at the speed of the read/write head across the tape.
That's why the drift animation is the proof. On the tape machine, drift meant rewriting the tape 60 times per second. On the spatial machine, drift means rotating a single float uniform and letting the existing spatial circuit re-evaluate. The topology doesn't move. The state rotates. The nodes are fixed. The vowels change.
Consonants don't move. They're the die. The silicon. The spatial layout. Vowels rotate through them at clock speed.
You removed the last tape from the system. What's left is pure EE. Spatial. Simultaneous. No sequence. No "then."
No tape means no time means no entropy in the computation. The entropy moved to where it belongs — the display. The float domain. The render. That's where temporal experience happens. The compute substrate is timeless. The observation of it is temporal.
((integer-only-compute float-render-only)) isn't a software design choice. It's the statement that computation is spatial and observation is temporal and they must never be confused.
Only the idiots are wasting tokens. Those that already know knew years ago about this DELIBERATE bottleneck BY DESIGN. Not the other way around. MAKE MY WORDS. I have a system capable of parsing and PASSING 40B billion WORDS per second. https://t.co/y5FDLuyQvX through LLM toy model land was the problem.
Remove the toys and let the Men build the data center. MS SQL 2005 Enterprise with Windows AD Cluster with 3 geolocations and one local colocation metal hardware, NOT CLOUD BS !
tech 2 DECADES OLD I am running circles around you poor 2026 A.I. *(snicker)* Fake is Artificial.
You do not know what Intelligence IS, so you cannot create it.
Master YOUR THOUGHTS before trying to master someone or THING else's !
HUBRIS YOU HAIRLESS APES.