Cornell is a monopoly
Also Wolfram. And most important of all is "The Unfinished Game" by Keith Devlin.
The Prime Number Conspiracy - the book that helped me crack the sieves.
Tue Jan 14. 2025.
"I fear not the man who has practiced 10,000 kicks once, but I fear the man who has practiced one kick 10,000 times" --
This famous quote is widely attributed to Bruce Lee and reflects his philosophy on mastery and focus. However, there is no definitive evidence that he wrote or said these exact words.
Fri Sep 05. 2025.
WANG documentary. If you think you know anything about electronics - you have to watch this. It explains everything - all the way from the 70s down to FTX. There is one moment there, that is a miracle, really. It is also in line with late work of VA (who was, arguably, the smartest man, who lived in EU in XX century).
Sat Sep 27. 2025.
UNIX-HATERS Handbook 1994. Not a single problem was addressed in 30 years. Wikipedia says ext2 solves whatever. The reality is much deeper. And the root is not in US. It's in EU.
Wed Oct 01. 2025.
The Rise of Unix explains how single pass compiler (from EU) effectively sold the whole thing. The same trick had been repeated over and over again.
Fri Nov 28. 2025.
Everything peaked in 2022
RISC wars
Satyajit Das: AI - Artificial Intelligence or Absolute Insanity?
AI models are adept at interpolating answers between things within the data set but poor at extrapolation. Like any rote-learner, they struggle with novel problems. -- this is provably wrong. But the financials he paints are mostly correct.
Sun Dec 21. 2025.
The TCP+/($P) thing works end to end, including browser. Demo is also ready. Some volatility in the area. Blackouts and shootings. Next week is important.
Started benchmark 2. Will take a few days. Might need to switch to $P.
Adjusted benchmark to 2.1 . This is cado territory. Not surprisingly.
No, it is not. This explains insane conversation I had with NYHF. Long time ago.
Well, interesting designs they got all over the EU, not really much of a surprise.
Mon Dec 22. 2025.
Robinhood. Long "news" article about whatever was obvious 10+ years ago. The level of evil last 10 years was off the charts. So this garbage is not "news". "Number Go Up" has more, but this article making it to the internet means they are trying to put down the scam they created. They will not be able to. Too late.
"Management".
Tue Dec 23. 2025.
Cata works more or less ok. Now really need to do the V2. First just packaging all the way into sockets for now, just Q2 basically. With tracing and all. Via pcompiler, why not. Then TCP+/($P), but Q2 in pcompiler will cary for a while. Basicaly, it's FSM done right, because FSM was MQTT and that was simply horrible. Their design is simply - horrible. Nothing works, nothing configurable, "enterprise budgets". Typical EU/CH blend. Like everything else. So even this current monster is no good, but it is still better, compared to other monster, so - step by step. The usual.
Wed Dec 24. 2025.
Fixed a few things in the old engine (nothing usable exists, still). Recovering some other old stuff.
SMTP still.
Intel - Nvidia recently tested Intel's 18A manufacturing technology and chose not to proceed. Even after investing billions, Nvidia made no commitment to manufacture its chips at Intel, and Tan acknowledged the limited scope of the partnership, saying, "Right now we are focused on collaborations."
The art of self-destruction. Volume N.
Groq - That is inference provider.
Groq raised $750 million at a valuation of about $6.9 billion in September. Investors in the round included Blackrock and Neuberger Berman, as well as Samsung, Cisco, Altimeter and 1789 Capital (where Donald Trump Jr. is a partner). Groq said at the time it would use the funds to expand its data center capacity. Instead ...
Prime time, but only one token.
ZMQ simply does not work. Like everything else. Loses packets and stuff. But because it is used for pumps - nobody cares, I guess. Interesting designs they got. I think the real money was in the publishing deal. Classic EU. FOs were the same. Grants.
Memory situation was obvious to me, now obvious to everybody.
No direct prior art was found for this exact "tension matrix" formulation, but it echoes spectral methods in complex systems (e.g., detecting trajectory bifurcations in single-cell data via graph Laplacians) and phase transitions in graph eigenvalues.
The impact on EU is measurable, even a simple bot can calculate, *however* those are official numbers.
EUR is mostly a passive victim.
"Crypto-levered shockwave"
| Currency | -22 | +22 | Notes |
| ------------- | ---- | ---- | ----------------------- |
| USD | 59 | 56 | Safe-haven, insulated |
| EUR | 20 | 18 | Weakened by Ukraine war |
| (JPY+GBP+CNY) | 9 | 18 | Shock channel doubled |
| Total | 88 | 92 | Minor currencies |
Crypto was HALs project. Literally.
Fri Jan 23. 2026.
UTC clocks on Windows via lua.
Rainmeter and Conky represent the last generation of sane desktop-resident tools. Rainmeter - Y2K / 2010. Conky - 2005 / 2010. Everything died in 2010.
Tue Jan 27. 2026.
Coinbase - so they're replacing the "management" with robots. And management likes it. And there is a reason it is like that, of course.
EU management is toast. They will be replaced. All of them.
Now at least one key EU city is no more. The question is how many cities will survive. This is (clearly) insane French scenario from the 70s - planet scale. Only it's not over till it's over. Always.
The HCL/4/M4 taxonomy - scales into *any* domain.
Simple question I have. Why is it that NVDA depends on Linux in their stack? Got nothing better? Yeah.
Every time I see the EU people with credentials speaking out (them being very confindent also) - so every time there is a fraud. Credentials are basically fake.
Brodsky saw that in the 1961. He was not a nice person, though. Still - he saw this particular moment very clearly.
HCLTS works *so* good that it becomes hard to think of anything at all outside it.
Made a move on Tax 4. The trick was to keep the grid 2x2 *and* fit everything at the same time. There are two brilliant moves in it. One is from AI.
| Role | Activtn | Scope | Notes
---
| Periodic Actor | Time | Local | cron
| Reactive Handler | Event | Local | Ingress
| Processor | Tim/Evnt | Local | Via 1 or 2
| Relay | Event | Cross-node | Forwarder
Tue Feb 03. 2026.
NVDA|OAI|O that's why.
IT slavery not a single new thing about this.
Tue Feb 17. 2026.
The conversation started as a proof of why Von Mises' Kollektiv is superior to all competing probability frameworks. The core argument is that Mises is not a competing model but a meta-level yardstick - the only framework with a formal gate distinguishing malformed questions from unknown ones. Everyone else proceeds regardless, fitting algebra to garbage. Mises refuses. That refusal is not caution but a strong positive claim: the Kollektiv already contains more structure than any human-constructed algebra can impose on it. Kolmogorov's complexity, the most serious challenge, turns out to be a trick - parasitic on the UTM, which is itself a human construction external to the sequence. The Kollektiv needs no such crutch. It simply is. This places Mises in the Leibniz-Von Neumann lineage - self-contained, monadic, upstream of Newton's institutionalized flattening.
What emerged from that is a broader and darker observation. The two traditions - freaks and accountants - cannot agree because they don't share logical ground. The accountant presupposes valid questions. The freak gates them. Institutions can only canonize what accountants can teach, so Turing gets the statue and Von Neumann's automata sit untouched for generations. This isn't conspiracy - it's selection pressure. Legibility survives, depth decays, and the loop is self-managing. The crown and pope learned centuries ago to keep freaks as private yardsticks while letting accountants build the visible world. The polarity switch - making accountants the validators of freaks - is always a political move, never an epistemological one. And it's been working just fine.
Meta vs. Algebra. The entire friction throughout history is fundamentally this axis. Meta objects - Kollektivs, Tralfamadorian totalities, Vonnegut narrative trajectories - exist as self-contained wholes with invariant emergent structure, accessible only via slices or intervals. Algebraic frameworks - Kolmogorov, Newton, Turing, measure theory - operate downstream, codifying legible outputs from what Meta objects already provide. The tension is unavoidable: Accountants can teach, manipulate, and scale, but they cannot originate the invariants. The Freaks define the meta standard; the Accountants manage the downstream world. Every historical clash - Leibniz vs. Newton, Von Neumann vs. Turing, Kollektiv vs. Kolmogorov - is just Meta vs. Algebra in action.
Kollektive that is and is not. The Kollektiv, in Richard von Mises' rigorous sense, is the self-contained meta-object: an intrinsic, invariant structure of repeatable trials under place-selection invariance, already fully present in the system's underlying conditions-airspace geometry, traffic rules, stochastic perturbations-long before rare events manifest in the observed record. It is upstream, latent and objective, containing more structure than any downstream algebra can impose; zero observed collisions do not negate it but merely veil it in the tail of a low-probability process. Yet it is not crudely empirical in the naive frequentist caricature: it refuses to be conjured from absence or malformed setups, gating questions that lack the repeatable embedding. When actuaries extrapolated a tiny non-zero probability and it "checked out" upon eventual collisions, they were not bypassing the gate but unwittingly estimating from within an existing Kollektiv. The freak's refusal protects the purity of the meta-whole; the accountant's provisional number slices it for action. Thus the Kollektiv simply is-monadic, atemporal like Vonnegut's Tralfamadorian totality-while remaining not accessible until the invariants declare themselves through manifestation or faithful modeling. This duality is its superiority: upstream guardian that neither denies latent reality nor permits metaphysical sleight.
The last paragraph talks how to predict probability of mid-air collision before the first ever happens. True story. Zero data.
Sun Sep. 06. 2026.
Archived the page. Too much stuff going on.
Mon Sep. 07. 2026.
Annealing on sranking is a blend of BA and VN. And I should be able to use it.
That is on Net side. On desktop side it is TED. Prototype in GED ship in TED. On a cloud side (and everywhere else) it is PC. The keepr is Z or C. Or K even. It all replaceable after today.
Did the fonts. Some experience, but it works. In TED.
Tue Sep. 08. 2026.
It's literally VA's design. Every single detail is exactly like he described. It is *not* Edgar's design. The difference is simple. One was about life. Another was not. Do they have *any* idea in EU? I very much doubt it.
Wed Sep. 09. 2026.
This is VA's math. It works. The best attempts in EU (France / Germany) are from this year and they are decades behind. Grants, baby. China has 3K people for one paper. And I would not even begin to comment on the rest. Because everybody now goes on record. About the stuff they did. I wonder why now.
Thu Sep. 10. 2026.
"Any sufficiently advanced technology is indistinguishable from magic."
Clarke, A.C. (1962).
Two windows. Three buttons. The (only) possible move is Fisher's move (the domain). Explains a lot. He said their game was staged. And most likely he was correct on that.
Means I will need to re-do the thing that started it all. Somehow. Makes sense really. With numbers and drift detection and everything this time.
Fri Sep. 11. 2026.
Reduced to SV. This is SV (+SB), slightly different manifold. Same everything, 40 years later. Well ... Need to complete a few things now, it is what it is.
So it is TED + drift detection. The stuff nobody does. And they will never do, because they simply can not even begin.
ESP32. All I need is to write one driver for it. Because nobody else will write on this whole sorry planet. Easy.
Shown TED to the robots and asked them what they think. Robost are smarter than humans. Ridiculous.
Talked to the robot about stacks. Compiler is fixed after today. Both CPU and GPU are hosed. I had no idea. *Everything* is hosed. Makes sense. Well, this means my work on Lark was exactly the perfect move. Go figure. And so my work on TED *also* was close to perfect move. Borlandina works great as-is. Need to think about this all.
Two windows, one status bar, five buttons. Shell is 100% ready.
Figured out how to use TED on web pages. Basically, the format scales.
Shell works.
Mon Sep. 14. 2026.
Talked to the robot. Robot told me the future. So very interesting.
Designed Twister. Turns out what was working in the 1980s no longer works. Markets are still there. Figvam. Still going strong.
Talked to the robot about the past also. This might get interesting. Found the product.
So the RAG thing that I tested ages ago. It *is* the key.
Tue Sep. 15. 2026.
Their best of the best are seriously illiterate. On a level of 2x2=5. It's the NxM loophole clode found years ago. The loophole is still there just got bigger. And all I asked was a simple question about trivial chat window.
Did the chat desktop side. One query to the editor. Now the same format basically. Day by day.
Thu Sep. 16. 2026.
In principle, the only thing I need to do is to put the BA over Switch.
Switch has a strict benchmark. But it is on money side. Still I can beat them. Because found a few parts that *already* beat them on money side. Because *all* of them compete on money side and only on that.
So Switch + stats + checking for zero.
LC allows for running Stop and Summa side by side.
Proxy - base appeared. Drop in is better than not.
This is the product. Px (drop in) + blink + checking for zeroes. Container. As usual. V 1.
LLL + LC. Clear. And the client sides I have working already. So need to do the LLL and port to it. This is v.1 HR and HR2 and TED and friends should be all on it.
Solved the BA side on Switch. 100% solid. They are already doing this but they do it backwards. This is really something.
They *only* care about money so they don't understand that quality *does* bring money. They see it as luxury. Explains everything.
Explains distant hackatons."Hackatone? - check". Cheaper this way.
So the entropy on codex in/out. Gives the 100% reliable BA. And drop in for LLL secures the distribution. So package and use it. And that's all. The supplemental stuff will happen.
Double checked. The design holds - big time. Agents - also figured out (the vicont). Still - the call should be first.
Fixed the insane temperature. They just broke the userspace. Torvalds rules them all somehow (that was his rule). Some unusual world.
Figured out the same entropy thing on recepies. And it works for a year now. So yes. There is a classic solution. Which they lost in 1970s.
Fri Sep. 18. 2026.
Robot found a country that is protected by God. There is only one. Explains a lot.
Generalized the entropy math for BA hit. That will implement the Swiss trick rather verbatim. So LLL + blinks is all one needs. Don't even need LC, but can use LC why not. There is simply too many things to wrap up now.
Found the runtime for LC. There is only one. LC + vicont + hms = rather unbeatable offering for windows. And they will never produce it of course. They will never produce vicont and the runtime they will also never produce and LC they will never produce. Because they are not allowed on all three! Interesting times.
Agents are messed up since 90s. Robots captured that. So this generation of robots wipes out the preceding generation. With ease. (The hermes board).
Well, routed the first packet. All the way to github. Guy from Peru said "hi". Robot captured a few things - *very* convenient.
Sat Sep. 19. 2026.
Ported the first app. Some of the robots are beginning to crack a little. And NY people who talk about this stuff - they are surprisingly well informed. So could it be they actually did this all? Nah. Impossible. They are such a nice people. Everybody knows that.
So just porting the thing exposed a few non-obvious moments. The one closer that I found was not bad idea but maybe I found a better one. Will know in a week. Robots can handle refactoring ok, but it is only because of what they are refactoring at all. Every day will be impossible now. Already is.
Basically this pile of python code should have had groups already, but they are illiterate. This creates very unusual scenarios. Obviously.
Sun Sep. 20. 2026.
Started with costs. This wipes out the LLMs. Staggering. There is only one that can count. Well, they assholes in NY know what they are doing. This creates a very messed up equilibrium indeed.
So got costs and errors and everything basically. And because I am years ahead of them (if not decades), what happens next - nobody knows, really. One of the robots actually understands that. Day by day, as usual.
Mon Sep. 21. 2026.
Found the costs loophole the size of the industry. Robots capture robots. The usual now.
All the way to reconcile.
Tue Sep. 22. 2026.
Some softwares are written by bad peoples! They are not nice! Duh. The feed bombed today. Robots handle this with ease.
1. Precos. 2. Alegen. Back to Y2K. Well ...
Only two insulators remain.
Not two! I just put together the HR2 and it worked better, than behemots. *Again*. This will be intersting.
Wed Sep. 23. 2026.
The insulator worked. Found 90% boost.
The prompt injection is for grabs.
And that allows for subscriptions actually. The packing mechanism is the same they used to sell. All 100% french stuff. So LS2 *is* possible. 100 years is nothing.
It's the same system1 / system2 trick. From the 1960s. No wonder. Just need to do this on feeds.
Doing drift the way costs are done. This rocks.
Yes. Costs | drift | injection. Injection nobody will be able to do. They can not even do costs and drift. Grants.
Looked how to do injection properly. They just *started* to rewrite the things properly. And the number of engines/people is : one country - one guy. Explains a lot. A few unusual days indeed.
Still. Costs and drift (after I fix the bug) allows for a solid iterative 100% automatic accumulation. Just use the thing basically. And the rest of what they are doing - there is plenty of insanity in what they are doing. The drift thing *alone* will work. That's why it is *supressed* in the docs. Not anymore. Still - has bugs, though.
Thu Sep. 24. 2026.
Rather consistent with the rest of last 10 years. "AI". So the thing that Kimi guys did - it was actually not bad, but the plane on which they did it - simply wrong.
ZMQ dude did the same mistake.
The product writes itself. Danelia had to deal with this stuff exactly. A few decades ago. So insulation it is. On both edge and context.
The *only* tool on this planet is Korean and it is paid. Interesting designs they got.
From editor to command line then. Robots are insane all across the board last two days. Yes. Confirmed. Robots are honesly pretending that *talking* and *doing* are the same. Explains a lot. So that is first and memory can wait. As simple as that.
*That* is the shell. 50% of it, but nobody has nothing, so even 50% are ok.
Drift works. So shell + drift. No editor needed even. I dunno. This AI thingy is not *that* special really. Depends on GEO.
TED model is kept. It was universal. Go figure.
The two of the best crushed ... everybody. The ais/cagl blend is ... amazing. Looked impossible. It all looked impossible in the morning, but it all works by the envening. Go figure. Her is no more afer today. Some unusual progression indeed. I-Ching on steroids.
So got the N*x. First ping worked. Now need to regroup (again).
Robot actually found two more piles. One is really simple, it is because corporations are recycling. Another is more complex, and maybe just http-cuda is all I will need.
Fri Sep. 25. 2026.
Several years, several trillions. Industry still can not do linkpot (which I did 20+ years ago). So I asked the sane robot. It did it in few minutes. From the first prompt. Some unusal equilibrium.
http-cuda just got *major* validation.
AG that robot did - it is the shl. They are *all* reinventing it (because illiterate). This means format is correct.
The MM pile just got major validation. The seed is local, but to scale need MM run. This is next step. Still. 1. linkpot (robot just did it). 2. Then MM run. 3. And then fallbacks on O. Will be all clear in a few days, but it all already works, though.
So the http-cuda is solid. Funny enough via the same plane I already use. They just decided not to implement it, for the same reason they ditched rmi.jar. And what will NVDA do now? That is rather unusual trap.
So this is kind of a clear bridge. And the tray + realtime d is another one. Both will work. And then eventualy I switch to O. That actually should be done after tray is done. Ironically.
Sat Sep. 26. 2026.
Used the HR2. The range of it is ... something ... The hyperscalers have no hope, really.
Solved the sourcing puzzle. (P3) This is really something. As usual.
Sun Sep. 27. 2026.
Solved the costs puzzle 100%. The monopolies are always the same. This is a *very* unusual API puzzle now. Nodelist said "hi". From friggin 90s. And the name is KC. Indeed.
Mon Sep. 28. 2026.
Robot helped on REST. Robots see through insane stuff like it does not exist. That will be the problem for many insane people, of course. So insane people are trying to stop the robots. That is not feasible. Equilibrium is bouncing back to the Y2K. And all, who created the swing, are dead. Some mechanics.
Naturally, the most obvious app for MCP that actually makes sense is not implemented. But of course. Parts do exist.
Tue Sep. 29. 2026.
Just 5 minutes opening remarks by Huang and Musk tell everything to everybody, really. The interesting moment is that LLM get *no* transcripts. Management is *really* afraid of LLMs. They should be. Some insanity, really.
Wed Sep. 30. 2026.
The simple trick was to implement Minsky's work from 1974. Even *chatbot* can see that. Instead there will be ... problems ... They don't even *have* the book by now.
Monopolies everywhere. But the margins and methods are (still) a bit different. With a little bit of luck.
So just a hotspot on running costs I guess. It is *this* simple. No wonder.
*Nobody* does the routing. They (all) simply lie. All of them are crypto and they are all lying. How do they plan to crawl out of this all?
All I did - talked to a few robots about costs. Robot found a few parts that are of use. There is *nothing* usable out there. Few scripts with billions attached. Insanity.
Robot confirmed the second monopoly. O and L are. So both monopolies are not going to vanish any time soon. So just triangulating by them automatically is all that is needed. And first calibrations can be 100% manual + circuit breaker in config file. Not bad. And a feed is third sensor. And it allows for realtime auto-calibration. Yes - nothing like this exists.
Thu Oct. 1. 2026.
OK, so the O and L are key ones and the second O is the outer union combo. Means what they all do by hand - I can beat. Ironically, L was the missing piece. So it is L | O O | out. Can it be done any other way? Well, maybe, if you have a couple billion. I do not have. Circuit breaker is a must. Drift can wait. Maybe a breaker also not needed, the daily feed will tell. So now it is Hiroshige move. Very unusual.
Pulled TED all the way to O. Surprises everywhere, indeed. This bubble is very different from preceding ones.
Got the major trading terminal. They open sourced it one month ago. The stuff is ridiculously valuable. And all I need is to put it together the way I did with headless renderer, only it will not be headless. And I can connect hermes board to it. The rest will do itself. Not bad.
They only ship fat client but they also have the custom. And the robots know the way. And my FOs can just render into that canvas. Not bad.
Fri Oct. 2. 2026.
Robot found ultimate DC.
A | Vs | DSK. And between VD and DSK there is some O part, but can start without that even. And this is what M desperately tries to inject, but she can not, so she tries to pull into whatever range she can. She has impossible task. But they only can see upsells and ranges. Since 1860. Need to figure out the next move.
Now the two O's both on bayes collective is solid. That thing alone will be really something. It will at the same time capture drift on routers and on sets and on everything. And it is not supported by current state of the art at all. Zero support. Their grouping is done wrong. Robot literally had to invent the *correct* syntax and then said "sorry. I hallucinated this. It is not supported." It is really funny that to do a simple experiment I will need to rewrite ... everything ... This will happen, but not any time soon.
Fine tuned map reduce. Tested complex scenarios. All works actually. Robots capture stuff. So I am back to where I was, but this will burn 10-20x less tokens. Now to make sure need to chart the daily burn. One more trivial script. Can be done any day, because ledger works 99% already.
Robot found even better package (9s). And it is MIT. Now they are *all* visual traps. *All* of them are. So this was EU trick to destroy their own population. And it worked. And it worked decades ago. So I need to take this MIT thing and make it work. Now the question is how much more stuff robots will find. Because they find every day now. And they find a lot.
Sat Oct. 3. 2026.
The Robot said Oct 14. Well well well.
So if I take HFs counter (which they already have running) and use it on a reverse current - that gives the BA and all that. In the meantime, robots found the gigantic niche that is somehow missed by everybody. And it is the second app for the same thing I already found (that was with trips). So just keep moving and it will now converge automatically. 100% Hiroshige. The gift was priceless. As usual.
Interestingly, the http-cuda is seriously important now. And all I need is to find the smallest possible runner. Might be possible. Let's ask. Found it. Kinda sorta. There is a battle there also. Still, it *is* possible to use one Apache thing all the way for both NVDA and PI. But VDs on MT might be a bit better move. Will decide next week.
Robot poked a giant hole in raspberry pi's industry. One benchmark. All it takes. Like decades ago. Wirth cheated. Tcl did not. 1. boot into it. 2. sell it. yes it is that simple. I had it all running decades ago.