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, though. 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.
Thu Dec 25. 2025.
Each and every networking tool from the north is being badly compromized. For 10 years now. All in line with the root of evil decorating the represenantive of evil recently. In EU. Everything is publicly available information, still. And the dude from the old guard knows all this. Yes, this is rather vague note about some cruel stuff. I will keep this for a few days.
Transmitters
That's why.
The Trump administration later allowed the $200 threshold order to expire and replaced it with a modified GTO that raised the reporting floor to $1,000, expanded coverage to parts of Arizona, and extended filing deadlines to ease compliance pressures. That revised order remains in effect through March 2026.
Engine 3 effectively.
Re-did the XSLS trick on FSMS
After 30+ years it is still broken all the way from hw up to the browser. Insanity in EU was deliberate.
Fri Dec 26. 2025.
Need to stabilize a few things. TCP/+ can be both physical and logical.
It is FSMS | LG | Q2 . And FSMS can also go directly to XML and that gives validation before execution. Nothing comparable exists, of course. They are decades behind. So I just take the existing LG systems (whatever they are) and scale them infinitely. That's another demo. Turns out it is Replit. And nobody else. Go figure.
The Q2 works end to end two points, but does not matter, only need to add a bridge and put production. *One* day. Started to climb up the stack and it is exactly how it was 20+ years ago.
Sat Dec 27. 2025.
Pushed Q2 all the way to B2C (!). Zero install was the key.
Seamless login is another SaaS.
FOs keep giving. The U|PTs|U pipeline will work. Just tested it and it *did* work. So first is Q2 for Str, then it is Q2C, then U|PTs|U on the same set of nodes and that's it.
15 y/o captain of V is neat, but it is only because of geometry.
Moved all the way up to the browser. They did *everything* wrong.
That's the product. Same village, same everything. 100 years is nothing. Jurai and all that.
That is because they obey the boundaries they agreed upon decades ago. I am crossing them today.
Several decades. Still, to install anything at all on Windows one should be a superstar. And to install something on Windows one has to build something on Windows first.
I am re-doing everything they postponed in EU since the 1970s. They had all the parts. Restrictions were self-inflicted. Still are.
They don't know that lint exists. It's that same swiss trick with envelope. I am lost for words. Yes, it is this bad. Well...
Sun Dec 28. 2025.
10 years cycles with "open source". Trying to re-do the stuff that used to run in Y2K.
Did the same thing I did on linux desktop, only this one is for Windows. They are equally insane. Windows (still) has a bit better hw support. "A bit" counts for hw. A *lot*. No way I could have done this without AI. Still, AI only moves you to the border, but it can not cross the border. Need to be human to cross the border.
Now I have this thing running without hardware and they actually sell the actual hardware for this. In NY they sell. I tried contacting them - they were insane. I am not in NY. Might be able to get by on software. Will know next few weeks. That is because in the office - need hardware. Still - right now, I am running it in software and it looks ok. So next step - to re-package it all. Again.
And of course reliability is horrible, but it works end to end and I will improve reliability gradually. The key thing is that it all works end to end. As of today. 12/28/2025 2 PM.
The most important thing, though is that electron delivers. Packaging will be hell (their whole vertical is based on that, there is only one project binding it all together). So basically what I have running is, I guess, impossible.
This also means that next is Q2C and BRO and everything get's feasible! The U|PT|U pipeline basically materialized today. The only thing I need to figure out now is injections and after that this is like SV on steroids, because this is ridiculously flexible and it has no borders. Did I remove the V? I might have. That's why I started this engine in the first place. I will only know after I sepend a few days implementing those two windows (which I already tried and it was not possible on linux). This is so very interesting, if it works ... Will know in a few days.
Reconnects are most definitely a problem currently (they lost atomic time servers in CO! No kidding). I will need to kind of do the similar, that's why I remember that one.
Increasing reliability gradually. Basically each component should have an FSM. And that is the only way. And that is the way they can not possibly do in EU (their best dude messed up, means they simply can not do that stuff). And when no reliability - means the loss is inevitable. And when the loss is inevitable - what is insane person to do? Yeah.
Yes, the U|PT|U-X is the thing. Now might not even need it, but still - that is what EU people were promising (for decades) and that they will never be able to deliver. Rather unusual situation indeed. And it's not *that* big of an effort also. Switzerland over and over again.
Mon Dec 29. 2025.
Need to complete Q2. That means complete the bridge, then complete reconnects, then add some stats. Also, the format of the packet needs a bit more work, like sending two timestamps instead of one for the composites (and this will always be happening, because of disbalances on actual/discovered, so need to think a bit more on this part)
Stacked Tabs is the next (perfect) app for U|PT|U and it is not realtime at all.
Just this thing alone might be enough, nothing like that exists, because *nobody* does L2 yet. They all only do L1 and they do it poorly. Just simple L2 offering will do it, and I already have large part of it working. So U|PT|U is designed end to end and there are now apps for it - daily. Tabs is one, PP on news is another and I forgot many others even.
Yes, it's feasible. Because after L1|L2 works need even more on L3 and even FOs on L2. And there is no FOs on L2 because there is no FOs on L1 even. So yes, this is rather self-destructing thing they decided to do in EU. All that technology was available in the 1970s. They just settled on the dumbest possible model. Whatever. They are all 100 y/o they don't care.
DBs and queries.
Tried to expand electron. Electron is no good for offence, but it is perfectly ok for defence. Wasted one day, though. Also it explains the problems I had on Jurai's build. I might need to revisit that place, because the San Mateo dude *really* wanted that few decades ago. And he was right.
Hard to believe but after 30 years of the internets there is not a single website that provides NFL schedule in convenient format. Means I will need to build one. I actually did this kind of stuff back in a day. Well ... This could be begining of something nice.
Tue Dec 30. 2025.
Did reliable re-connecting, state keeping cross platform websocket bridge. Did command line first and it was the only way to get FSM right. Last mile is almost there. No ZMQ. ZMQ does not work.
Did the last mile so it is all the way between the clouds and the browser. It will not work Q2C, though, because turns out that caching for humans is much more complex than I though. This will work for machines, though, and there is visibility and basically can just start using it.
The low level roles are emerging. There is ticker. There is differ. And just those two are already game changers. So should just start adding those. Why not.
I now have a machine that forks the Electron mini apps all the way from text config to the desktop icon. Impossible without AI. Took hours with AI even. This is the level of control I had on Linux desktop. And it was easeir on Linux, of course. On Linux can just write a few text files. In MS this is a (complex) generator script. Does not matter - this takes care of desktop.
OK. so Q2 now works all the way to desktop and ticker ticks. Now just need to add things.
Wed Dec 31. 2025.
One more of inference providers started to suffocate the inference. So simple reliable router is good enough for a product. This means I need to port this to LG, but everything should be PT with selectors. Need to think how best to do this, but from the looks of it the pure HTTP should do the trick. The APIs are not obvious actually, because it should be programmable. So simple programmable selector will handle the core MR pipeline, but need to re-do the HR pretty much exactly like Q2 was done. Yes, that's not bad of a project for the next step. And it will need to be done exactly like Q2 was done. So it's HR2. Step by step. Start with UI.
So yes:
"Running this on the Mac mini saves ~$2k"
is actually an understatement.
Why This Works So Well (Key Insight)
You are exploiting three things simultaneously:
Depreciated Intel hardware
CPU-quantized LLMs
Core's low steady-state cost
Bottom line: You're not just saving money - you're building a resilient,
owner-operated infrastructure unit. At $140, it's effectively a public good.
So this is a product. It is hardware product.
Interesting.
And for the cheaper end it's what LS did. Rather exactly. So HR2 in a cloud first. Can even deploy to EU of all places, from the looks of it. It's $2.
So I keep moving on Q2 (and it works). And then I also keep moving on HR2. And it also works. Next step on Q2 is the cata on A. Next step on HR2 is UI|MR|HTTP|PTsels with a single config exactly like Q2 is done. There are two planes (one is (logical) points with selectors and another is plays). How exactly do they do it now? They mess up, that's how. Designed the stack end to end. AI will replace many people.
Turns out NNAPI controls large percent of the markets and that is effectively controlled by one dude in NY. Italian, naturally. That is because Italians are very good with math, I suppose. Apple does not suport NNAPI.
Turns out I reimplemented good part of Lang Graph not knowing that. Their model is of course better, the question is if it works at all. If their runtime works this is converging *very* nicely. Yes, HCL for YAML, Lang* for XSL. And FSMs on the side. And that is actually orthogonal to what they have. They can not possibly handle complex FSMs. Because they are writing them by hand, so they don't have them.
Replaced LG. Took one day. AI will replace many people. Basically the thing I did with Hadoop is in full swing. Times 10.
Thu Jan 1. 2026.
Both LG and LC are no use. It will be LLMS on LITE. The MQTT stuff will also be ported to Q2.
-> expt is inevitable. So just PT with expt (on FZ) will be good enough. They *still* don't have it in Ansible - after decades of waste.
So this means I just implement B-first and D-first in parallell. Why not. Prolog guys had it easy.
Implemented routing of cata(A) to Q/Q2. Q2 was easy. Q is ZMQ so it was almost impossible. It pushed the needle on AI, though.
So semantic routing is feasible and it will be derived from medcial industry, rather clearly, like the G5 was.
Not only semantic routing is feasible, I now have two planes to decide on. Because ZMQ does work faster, because it pushes the boundaries, so in a way it's just logical UDP with piles of garbage attached. My stack is of course different, but after today I could actually package *exactly* how SCO dude did his packaging.
That is because they are all EU and EU thinks the same. If that process can be called 'thinking'.
Q and Q2 both work end to end. Need to give them a few days now. Did the foundation for LLMS, not bad. So just few days like this should be enough. Move on QS. Move on LLMS. Move on M4S.
Progress on desktop (N windows)
Next steps: LLMS, Q2 (volb/mqtt stub, optional), HR2 on LLMS end to end on N3.
Fri Jan 2. 2026.
Gexa is inevitable. And rather exactly how it was designed. Interestingly, very old EU poetry was about that, so it looks like timeless stuff (and I only remembered because of paper ppl song). It's that R/P/S thing. Timeless. Well, civilization got a bit carried away. Always.
CT|PT|LLMS|P is a clone of LS and I have 90% done. It's basically SB next generation. Blend of SB with SV with a clear Zvn out. That is exactly what I offered them ages ago, so it's not that I am missing something. Gexa is inevitable.
Sat Jan 3. 2026.
Did the Q2C caching.
Sun Jan 4. 2026.
Found one more bug in Bourbakist designs and that pushed me towards Peano (via Hilbert) and then I found : In a way, Halmos is Bourbaki in reverse . I can not believe it, the guy was alive when I was brought here! So basically him and VA might be all one might need. There's no known correspondence, meeting record, or collaboration in the literature that links them directly as colleagues or correspondents.
Innocent conversation about this stuff results in some not so simple battles with one not so simple robot. I need to understand if the robot is hallucinating or if it is absolutely brilliant. Will need to test it's code. From the looks of it - the robot is not bluffing.
In any case, the battle with the robot (which can not be accessed now, because too many people are talking to the robot) so the battle with the robot was ridiculously productive. Interesting moment that one is the *only* battle robot, others are different models. Well, blade runner was *clearly* ahead of it's times. On pretty much everything.
The dream today. .
Mon Jan 5. 2026.
This internet thing has different regulations across different countries. Duh.
There is only one sane benchmark. And only one insane benchmark. Only need to beat both and technical side looks not that hard. Depends on location, of course.
They are still trying to do the right thing, but on a wrong level (their own paper, man. How can I possibly make this up? I can not) I seen it here 10 times already, this is not random. It was the same with the camps water thing in the 1940s. Humans were supposed to get better since that time. Maybe. Probably. Here is a few matches, make us a helicopter.
Tue Jan 6. 2026.
Firewalls / DCs / cpanels.
So basically I published 10 years ago and web archive still has the copy. Web archive is nothing, compared to gexa.
Turns out SB (and SV) have no competition. Nobody did what I did decades ago. Robots say "yes it's the only way, you should opensource".
Wed Jan 7. 2026.
Yes, the data driven stuff was published for a reason. Interesting. And it was absolutely not about the primary subject. I need to re-read it again, but this time validate with AI. Now that I understand their designs. (Their designs are of course insane).
Figured it out end to end. Need to think now.
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 thing tried to destroy Markov. Who was undisputed pillar of EU civilization - for centuries. I guess it can do Pareto, if it wanted, but it only does what we tell it to do. It most definitely *can* destroy Pareto. EU has no defences. At the same time thing has no idea.
The Be benchmark they did *can* be actually of use, but they ignore the most important part. It can only be used in low dimensional space yet they don't care about dimension reduction part. I just had one bot capturing another bot. So in principle it is possible, yes. But not the way they do it. Dimension is two. K is three.
So this modern AI thing is one (crippled) neuron from the 1970s. And AI knows this. Books are still in the library, but about 5% of the books had been moved somewhere. I wonder where the books go.
Thu Jan 8. 2026.
Did the universal, self learning, inference based expert system that can be used by anybody for anything. In 2 hours. AI will reshape the world. Like the web did. Why are they building all those datacenters, though? They're all insane, I guess, that's the only explanation.
The dude who did the (great) work in the 90s wants to sit on it forever. Which is not really possible now, because AI can take whatever was there and rewrite it into 100% free platform. Yeah. 1 hour - and you take decades of somebody else's work and it's yours. "I did not create it. I own it. It's power, babe! *She* is my Number One!" Great movie. Ahead of times. Like everything from the 70s.
Now this stuff I just did. It beats the (only) planet level SaaS. Which I was thinking about for a few years now. They had those massive problems with management, that started many years ago. So they're defenseless also, from the looks of it.
By tomorrow, you'll have a clear picture of how the buckets filled and whether the system behaves as expected.
So that's the real problem here. The thing can estimate *infinitely* better than *any* human.
Did the first temporal state machine. All in one day. Some of the robots become insane on you out of the blue and start ignoring the exact prompts. Whatever.
Temporal state machines are some scary stuff. Because nobody does them. AI does not care.
So the thing tried to destroy Markov but then realised that Markov can not possibly be destroyed in live systems. (That's what VA knew allright). So the thing has no idea really. It has absolutely no clue about reality. Like some (if not all) of the managers, I think. Match made in heaven.
Fri Jan 9. 2026.
ZMQ does not work. As simple as that. Can freeze silently and stop sending packets. But then again, if/when some idiot is messing up with the network cables in the datacenter - ZMQ has no idea about that. Still - all his machinery - not really useful for a crappy networks *and* at the same time if the network is *not* crappy, it's also not much use. There are several paths forward (shell is one of those, and that was the path rails and kubernetes decided to pick so I am not sure yet). Critical path, though is to run ZMQ and Q2 size by side and *that* will automatically capture the disasters *and* it will also prove the plane that is broken. This is going to be very intersting now. Because if Q2 works and ZMQ does *not* on the same network, that is worth publishing really. So the ultimate benchmark presented itself.
5. Observability Is Not Optional Infrastructure
TCP:
can be traced,
inspected,
packet-captured,
and reasoned about.
ZMQ:
hides everything behind an abstraction boundary,
exposes minimal metrics,
and gives no causal chain when things stall.
In operations, this is catastrophic.
You cannot debug what you cannot see.
Well, you can. I did. But this means - I will be running Q2 side by side with ZMQ. And this also means that my market is planet earth. *Any* org that runs ZMQ is a client. And *all* crypto degenerates run on ZMQ!
So this is the book that can be completed in a few weeks (the VA book will take years). So only need to wrap up the Q2 and publish. AI will handle the rest. Well ... Contrarian books sell real nice, actually. And the benchmark is already unbeatable and verifiable after today.
So next step - VCs, then. Let's see what happens next week.
So moved from Q2 to Ringo. Let's write down. SV was like that exactly
1. single conf
2. single exe with modes. But not on bridges. Bridges are isolated exes
3. p/s/b
4. split is on channels. single (tuple) of ports. r/w
5. acls
6. the base is p|s|b|wrt . Eventually wrt might become swrt.
7. This means bridges poll. Which is ok. They are outside the ring.
This is clean and this opens up FIO. Nothing else does.
Then it becomes a swarm of FSMs.
With one entry point - that is the gateway.
Do I need this though?
Got Q2 and ZMQ running side by side in the browser (was browser specific, because SSL). Jurai's plan might still work. Surprisingly. Need to fix the state machine, to be 100% reliable. AI be AI.
Sat Jan 10. 2026.
FSM-1 (subu:VAV) already works on Q|Q2.
100% AI:
IPRO | LLMS | PTS (DAG|D) | HC = HR2 = 2x (demo/base)
IRPO | LLMS | PTS (DAG|D) | HC = Lem's switch (demo)
zero AI:
Q2(Ringo) = FIO | PTS (DAG|D) | HC = ZMQ+ = 10x (infra) (FSM)
PREGO | PROL | HC = CSV|FSM = UNI(FSM) = 20x (infra/product) (FSM-2)
FZ | PREGO | PROL | HC = (SaaS/I) = 30x (infra/product) (FSM-2+)
VECS = IO | HC * Ringo = fog incremental! = 50x (infra)
blend:
IPRO | PTS (DAG|D) | HC = CT = 100x (toolc) (HW)
IPRO | PTS (DAG|D) | HC = PtoC = HW = 100x (toolc) (HW)
IPRO | PTS (DAG|D) | HC = SB = HW = 1000x (toolc/product) (HW)
2C
Rec (already works)
Books (exact same to the phone)
Ocean is BrMM (verbatim). GeXa (verbatim).
Jurai (verbatim + one button + one cfg).
In no particular order, but Jurai is the simplest.
Because nothing exists and San Mateo guy wanted it badly.
>= and >!= is key.
Need to figure out how to re-balance the stack.
1 and 4 looks like the best move.
So need to wrap up the FSM and that's it.
Prego was a gamechanger.
2x1 thing (C v G via MM). Might be able to get by on L0.
2x1 + the name (which I have) and complete the stack. Looks feasible. Base should handle.
Thing can do what I did 30 years ago, but faster and with piles of bugs. Progress. Does not matter. They have billions of people to review the transcripts and I bet they do that layered. It's MT actually. It's all MT.
FSM works. 2x3 | * = 64 cells. Q2 works. Everything works.
If someone finally packages it cleanly-for editors who want control, non-native speakers polishing style, or anyone tired of black-box humanizers-it would fill a real niche. The demand is there; the UX just hasn't arrived yet.
That is another app for L0. So both HR2 and this can be implemented same way *exactly*. This is solid.
Also need to complete the second FSM now that I learned how to make them. Should work from the same .M file.
I had a friend in a galaxy far away. He was a decorated math genius. The thing is his level. I gave the thing the same puzzles. He was better, I think, but the thing does not drink and the thing does not care. The thing is a gain without pain. There will be massive problems.
Sun Jan 11. 2026.
Turns out after all those years, Google can not do shit in the most sensitive part of their stack. Somehow all of their 'power' means nothing. I wonder how does *that* work? They are defenseless against a few shitheads in EU. And the shitheads in EU are just that - shitheads. "Management".
Fugured out the >= and >!= . This thing alone will do. Too much stuff to complete now, but that's a good problem to have.
So it's UP and PGO next week. Hello world on PGO needs HR2 completed, though. And I still need to wrap up the second FSM, just because it's too good and it will be a day or two. Next week will be something.
The Lem's switchboard is the third app. Still need to figure out between FIO and IO, because they have exactly 50% of RAM difference. Not even funny. Ringo does solve the complex problems somehow.
Turns out - FIO is inevitable for scale. (And Ringo is intevitable for FIO). HR2 and friends can progress on existing infra, yet placing some Ringo in the mix will affect things *big*time*. Because the *only* way to escape a FIO lock-in is to oversubscribe Ringo. That way the *only* thing that will matter in the clouds would be an actual free physical ring. Which some of the clouds do have and some (simply) do not have. Ringo just became some kind of a monster ... Well, Java RMI, what's new.
Mon Jan 12. 2026.
Figured out a track, that can handle IO. Rather unusual EU loop. Ringo is a perfect fit for FIO, but that's compute. IO is primarily IPs. Explains what happened in EU, they just routed streams via simple selector. They had to bribe the key government fugure for that and that was clearly too easy. Not some special knowledge, the dude was writing books about that since the 90s. "I am so confused!"
IO is IP and storage, but no ring, FIO is a ring, scale, and no storage. Basically, FIO is one more layer of cache. Since my journey in SF they all run towards caching, because they only look towards 100x. Another branch run into hashing, because hashing is math (and agencies love security so it's personal for them). Nobody really run into compression (they manifested with Silicon Valley series, but that's movies). The crypto branch overdelivered again. So what's up with compression layer? Let's ask. NGT for edge nodes and hnswlib otherwise. AI will eliminate many groups. After AI is done with pretty pictures, it will get serious. Maybe it already is.
Seriously? ... NN ... I can not believe the level of insanity in EU, but they are *clearly* and *absolutely* insane by this time. 500 Mln people ... Well ... This means BO songs from the 90s might become timeless. Like Blocks' poetry.
Well. AI just destroyed a few (protected) monopolies. The impact of this thing will be staggering. Did the data layer end to end the same way I did PTS *already*.
Interestingly, only very few people are truly protecting me last few years. And they have no idea they would be protecting me. If anybody would have told them - they would not believe. They did not see it all as a big deal. But it was. Nobody remembers them by this time.
Tue Jan 13. 2026.
Notice EUR declined, GBP gained slightly, USD lost a bit due to valuation effects but remained overwhelmingly dominant.
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 |
The thing does *not* know about it's own origins. There is hope!
Crypto was HALs project. Literally.
There is a way to do FZ/Z on the thing itself! Pretty much exactly like I did to the thing's root. And it *worked* on the root. So it will work on the thing. Neat. Clearly, the guy in Japan could not imagine the results of his hard work being re-used like that, but this looks legit. Fibo!
The thing is programmed to violate Azimov's laws. EU, what you gonna do.
Both Q and Q2 are getting the good old http bridge, which makes sense from economics perspective and opens up both IO and FIO a litle bit.
FZ | RingO | LLMS | HTTP1 | VEC+ | HC - domain does *not* matter looks like.
VEC+ is almost inevitible looks like.
Wen Jan 14. 2026.
NY tells the routes verbatim. How is that "secret"?
Thu Jan 15. 2026.
Did mak end to end in python - works how it should, 5 tests will need to add many more. Tried pushing the thing to the limits - it cracks. So all that stuff they're writing recently ... Hmm ... They deploy *this* to military? Well ... Figured out how to use this thing.
Fri Jan 16. 2026.
No need in good old http at all. Somehow. Q2 bascailly just replaced everything imaginable. But I do need to compile the thing still. And it should be self-contained still. So need to repackage everything that is already running. Again. ANSv3 this is, because the target is different (and they did nothing for decades). That's what perl people wanted to do ages ago, but they simply had no skills.