Notes on AI
Page re-factored (again) Nov 03. 2024.
AI Tribes
- Symbolists. Ideas from philosophy, psychology and logic. Inverse Deduction
- Connectionists. Neural networks. Backpropagation
- Evolutionaries. Simulation is the king. Genetic Programming
- Bayesians. Everything is probabilities. Statistics. Bayesian Inference
- Analogizers. Detection of metapatterns. Support Vector Machines
Machine Learning
Pattern recognition, statistical modeling, data mining, knowledge discovery, predictive analytics, data science, adaptive systems, self-organizing systems. ML *is* subset of AI.
Sep 07. 2024.
Used AI on Barabasi + PT net blend. Amazing results.
Sep 15. 2024.
Well, let's write maximas here. Bayes contradicts (parts of) Number Theory. I have good enough proof for this one, had it for years, but not 100% formal. Eventually will write down formal. If time permits.Jan 15 2023.
This used to be a page discussing some stuff from this book.
The Theory That Would Not Die: How Bayes' Rule Cracked the Enigma Code, Hunted Down Russian Submarines, and Emerged Triumphant from Two Centuries of Controversy
The aspects were mostly mathematics, mostly Bayes, a little bit of number theory.
Then it got to string theory and ZK, so I decided to remove the page. I am in USA. SFBA.
There are plenty of other pages on this website, that are more conventional.
Jan 19 2023.
Some people say it's about expansion.
Some people say it's about perfection.
Some people pretend the choice does not matter.
It does matter.
How Redditors Exposed The Stock Market | The Problem With Jon Stewart - has a nice diagram(s) (and overall good explanation of what was happening).
Infinitesimal: How a Dangerous Mathematical Theory Shaped the Modern World.
Do only what only you can do - Dijkstra
Their book is frequentist book
Cornell is a monopoly
Also Wolfram. And most important of all is "The Unfinished Game" by Keith Devlin.
Apr 10 2023.
Markov *and* Bayes > Bayes ( but Fourier still holds somehow )
Sep 22 2024.
Basically insane bayesian was given infinite moneys and no stop loss, gambled substantial pile of EU moneys away, put boyfriend in jail. Walked. End of story. Not sustainable one bit.
Sep 25 2024.
1. Black box (of course, because valuable). 2. Bayes inside. 3. Margin call. So much bullshit around such a trivial math. Called civilization. You've seen nothing yet, basically.
Oct 02 2024.
Breakthrough on primes conjectures. Basically, explains why that German dude only worked on N, saying that it's the only way to do math right. Germany before the WW2 was an interesting place indeed.
Oct 11 2024.
From Perl page
So first they map Pascal to UCSD Pascal (San Diego 1977). Nobody cares about that one. Then after about 10 years later they map Concurrent Pascal to Emerald (in 1983, in Seattle) and it fails. Then they move it up North, and Canadian Gosling finally produces Java (which is some implementation of UCSD Pascal) in 1995. Then Java says that all you basically need is simple RMI, and pretty much all is needed for RMI to take over the Internet is for MS to include Java RMI with their MSIE browser. But they *already*failed* with Emerald in 1983. *Both* Emerald and Java are just rewrites of UCSD Pascal (from 1977) only one rewrite was done in Canada (and is called Java) and another rewrite was done in Seattle (and is called Emerald). Looks like MS could not allow others succeed in the area they failed, so they pulled the plug on Java RMI in MSIE (and eventually Sun is acquired by Oracle and that's it).
Oct 11 2024.
To think about it, I only got right one of the pillars of semiotics trifecta. And Don got right another pillar, and he did it 20+ years before I even looked at semiotics. The most interesting thing is that this stuff is ridiculously easy to explain and it exists for centuries. Yet most of the people would not even understand the importance. Well, Arnold figured it all out at one point. Very long time ago. That's why he started to make typos here and there. And the Void Summits dude not only figured it out, but wrote the whole book about it. Only took me 40 years to re-read the exact same book I was *already* given 40 years ago. 40 fucking years. Yeah. I managed to see absolutely no use in the book that might become maybe *the* decisive book in my life. Youth. What you gonna do.
Oct 13 2024.
The role of some countries *never* changes. The same country that managed to sidetrack WEB1 bubble is sidetracking NN bubble. The results are unusual in EU. Explains blocking of youtube across the board.
Took me 15 minutes with AI to prove that Moore and Meely are corner case of Markov. Switzerland all the way. Well, well, well ... Meely > Moore. Interesting.
Oct 14 2024.
Designed the next step end to end. For a key component I have to re-implement my own part from 15+ years ago, because ... well ... Yeah. Only I can do this in one day now (verus one year it had taken last time). Von Newmann said "hi". Like he already said to countless generations of idiots here. So did Turing. So did Wiener. List goes on. Of Markov / Moore / Meely - Moore is the most primitive thingy. Basically it can not possily get more primitive, than that.
Oct 16 2024.
Not what I was expecting, but let's consider the first test a OK. Need to rewrite plenty. It is interesting, how frequentists think they can possibly capture inference like "sure thing". Yeah, right. What they *can* do is to promise that to old ladies, knowing that old ladies would not understand a thing.
Oct 17 2024.
Major deliverable. Next milestone is 3 weeks from now. Tuesday, November 5 is the next milestone. Duh.
Oct 22 2024.
Semiotics
Oct 25 2024.
Umberto Eco's semiotics.
Oct 26 2024.
1. Yes, it is possible to implement his design. He is dead, so he would not care.
Oct 27 2024.
2. This crypto thingy. Clearly - this is just randomfisher(FX). Well, gotta give it to the guy. He was briliant.
Vatican representative announced Holy War today, basically.
Oct 28 2024.
Thank you Umberto Eco. Your approach to semiotics * DDVT totally works and it allows to close on several things that were bugging me for decades. DSL and Data. Eco * DDVT beats roadsigns. Basically this looks like the end to several decades of battles. Even a little progress on those issues already spanned a few things that spread all over the planet, such as JSON and/or YAML (and DL and TOML, which is DL also). Mr Umberto Eco, you're the best and your mistakes are also the best. It was only because of your little disbalance I was able to put it all together. The continuation of Peirce trajectory is *really* something. I need to double check a few things now.
Oct 29 2024.
Started to apply Eco's approach to reality. Published 1975! Lisp - Originally specified in the late 1950s, it is the second-oldest high-level programming language still in common use, after Fortran. UNIX : development started in 1969. Lisp | UNIX | Semiotics. Game, set, match. It was all over in 1975.
Oct 30 2024.
Markov keeps giving. 100 y/o maths. Well, Trump wants to go back to 19-th century. The shoe fits.
Nov 01 2024.
I am pushing LLMs to their limits by the exact same methods they used to push people to their limits here. 30 years ago. Simak wrote a whole series of books about this.
A good theory is:
Accurate - empirically adequate with experimentation and observation Consistent - internally consistent, but also externally consistent with other theories Broad Scope - a theory's consequences should extend beyond that which it was initially designed to explain Simple - the simplest explanation, principally similar to Occam's razor Fruitful - a theory should disclose new phenomena or new relationships among phenomena
Nov 05 2024.
Placed the first end to end pipeline production. Everything works. Kind of.
Maintenance will be a bit of a problem, because no tooling exists.
Nov 06 2024.
1. The work of Eco is absolutely perfectly applicable to computing. Why nobody did it for such a long time? The application of Pierce's step resulted in major and obvious advantages. Yet *nobody* even *tried* to apply Eco's step? The only explanation I see - the EU decided to go insane in 70s, they were too busy with that direction. That's the only explanation possible. They are doubling down basically. Too bad.
2. So I made major progress on (1) but now there is (2). That is a component that I *already* produced here, in US, by myself and for myself, in Y2K, some parts of it were stolen, the only dude on the planet who even understood what I was doing was French. In EU. There was another (Dutch) dude but he did not get it. There were also few British dudes, they were mostly neutral cheerleaders (and I was lucky they existed). Now by this time British are destroyed, Dutch don't give a damn. French are very unlikely to produce a replacement. There was one Israeli - he is not operational for a decade at least and his name had been removed from the Internet. Rather bleak picture. The component is *still* not complete. I will need to complete it myself. It's inevitible for 20+ years now. Nobody did nothing comparable.
Nov 08 2024.
Turns out that after figuring out the Italian part of the stack, the next natural move is to blend the remaining parts of Plan 9 with recent British (+Dutch) projects. Storage. W3C' vs 1C.
Most of the time I write stuff here - it is after I tested something and got results.
Nov 09 2024.
Infinite number of puzzles (literally). Relational / hierarchical is one of those. The pragmatic solutions are always non-linear. I once had to design the FPU, that's how I know. A prominent billionaire failed that same test. Publicly available info.
So this collision between Moore / Meely. Both are corner cases of Markov, yes. Moore is the most primitive imaginable. Yes. What's the role of Meely then? At first glance there is none. Only turns out it can be used to model Catastrophe theory. Markov is arbitrage.
Market cap of Dropbox: $8 Bln. Market cap of Box.Net: $4 Bln. Let that sink in. Market cap of Google is $2 Trln. Facebook $1.5 Trln. Cisco is $230 Bln. Oracle - $500 Bln. IBM: $200 Bln. Nvidia $3.6 Trln. Apple $3.3 Tlrn. Wells Fargo is $200 Bln. Disney - $200 Bln. OpenAI: $150 Bln. Interesting how this silicon valley thing works. Amazon : $2 Trln.
Nov 10 2024.
Let's write an elementary school thing here. No amount of money can buy the most precious object. The most precious object for EU is not the same as it is in some other places. When thief can not buy the most precious object, he tries to steal it or to take it by force. And that was the *only* thing EU idiots were doing for as long as I rememeber. And that goes back 40+ years. That's how far I rememeber. I seen it and I still see it. Bunch of loonies. Who lost everything decades ago, but can not let go.
Nov 11 2024.
Not every day one finds the logical bug that is embedded into a culture that holds mighty empire together for centuries. Empire based on a bug. Self-collapsing by dialectics. EU all the way. But all the way to the point. Explains the poetry that came next. Pazolini had a verbatim statement on that. Killed 1975. Of course. 1970s Italy. Everything happened there (and maybe *only* there). Well, the battle goes since 17 century, naturally. Like nothing changed.
Nov 14 2024.
Figured out a very nice BA/PT number. It's 8.
A bit more validation on M/M. Fundamentals work. Big time.
Nov 16 2024.
Well, F(x) -> M/M (* m) . Spectra stuff. Even a baby step goes a long way.
Even better. F(x) | StringsT | Markov(1/2) | M/M | on a Sieve(Y).
Could actually use each part separately, like everybody does.
Only in reality somehow *better not* (which is Markov again). The dude somehow holds everything together. For 100 years! The interesting moment here is that M/M * M is not N, but the pipe is almost entirely N. (M | M/M) part definitely becomes N. The German WW2 war dude was right about N. And that is really ... unexpected ... The pipe is almost entirely N, but some *parts* are not N and it is possible to have core N and the rest not N. In a way it is possible to shift/rebalance complexity between components. Taleb's "Fooled By Randomness" again.
Nov 17 2024.
No pipelines in Windows.
Looks like the German WW2 dude was correct after all. If you can not figure out the N component, most likely you're distracted from the reality of the process. Interesting and unexpected. In a way that's why brilliant Mark Twain lost all he got. Like many other brilliant writers in US did. Only they are immortal, in a way we are not. Back to Imaginarium again. Some interesting movie that was. Year 2009, yeah. Reality is - *have* to have both N and non-N parts. And that destroys several decades of Talebs' hard work. In a way, wrong ending somehow un-does good beginning. They know that in EU.
Now after the complex pipe is figured out, can slice it into (much) simpler parts and focus on tivial stuff. Which is not really trivial now, because the world was frozen for 10 years. And computers tell people to go and kill themselves. Because some millions/billions people trust computers too much.
Nov 19 2024.
Rockets in EU. "Madness and stupidity".
A few more engines. Usable ones are *only* in Scandinavia somehow. The trolls were staking the engines in Scandinavia.
Nov 20 2024.
So that thing I found in Scandinavia (one of a kind, planet scale). It can be attached to pretty much anything, that exists. Ironically, 10+ years ago that was done, and *that* add-on became planet scale business. Now I can do the same (better), but I can not do nothing at the same time. Well, what you gonna do. Idiots and computers make a strong force, indeed. Will know some results in a week. EU escalations again.