Umělá inteligence nebo přirozená blbost?

I apologize for the tabloid title, but I think it is time to open this "pandora's box" a little bit here in the following areas:


- artificial intelligence, resp. The technical means capable to some extent to "understand" the written word and somehow transform it have advanced to such a degree of proficiency, that I am considering how to use them in our environment


- the goal is not and will not be for some time to have it write articles and texts for us, please don't. I'm assuming that all the language models that are used in the background of these technical environments have "limited knowledge", it's not that the thing can invent anything, it's not. It can rephrase something it has already "read" somewhere, just write it in its own words (and sometimes not even that, it just copies a piece of the original text from its source, from its memory). In the case of programming languages, for example, it is capable of certain "learning" or "abstraction" and thus delivering the requested code, even if no such code exists in the innards of the machine, but of course this does not apply to history, which simply cannot be invented. And the second thing is that it is RELATIVELY accurate, but not HUNDRED PERCENT accurate, and that is a significant difference. And we know very well ourselves that it is considerably more work to find errors in someone else's text than to, er, make them up ourselves


- the areas in which I think this could help us are currently two:
- the recommendation at the end of the article for "similar articles". This is now only governed by the fact that it suggests articles in the same category, so if you read an article in the first world war section, it will suggest other articles in the first world war section. Nothing else. But the machine has the ability to somehow "describe" an article by a system of numbers, a vector that has about a hundred parameters, and by that to find out how similar or different our other articles are. Thus, he can recommend an article about a pilot from the Second World War, when we read an article about a pilot from the First World War, rather than recommend an article about chemical weapons from the First World War, because the similarity is "not so similar".


a brief demonstration, I took nine annotations of our articles and put them into the system and had it "rate" which are more similar and which are less similar:


entry:

1. Jugurtha was a Numidian king. When the Numidian king Micipsa, who adopted Jugurtha, died in 118 BC, Jugurtha and his two adopted brothers Hiempsal and Adherbal took his place. Jugurtha arranged for Hiempsal to be killed, and after a civil war in 112 BC he defeated and killed Adherbal.


2. The ASCOD (Austrian Spanish Cooperation Development) family of armoured fighting vehicles is the result of a cooperation agreement between the Austrian company Steyr-Daimler-Puch AG and the Spanish company General Dynamics Santa Bárbara Sistemas (both companies are now divisions of General Dynamics European Land Systems). In 2004, an upgraded variant of the ASCOD 2 was introduced into the armament of the British and Philippine armies.


3. Vehicles of the Navistar International 7000-MV type series use the same chassis as civilian vehicles of the Navistar International WorkStar type series. They have been in production since 2005 and are used by a number of countries around the world, including the US and Canada (under the MSVS - Medium Support Vehicle System project).


4. An extremely interesting use of the rare Sd. Kfz. 234/2 by the post-war Czechoslovak army


5. In 1944, the young Mandela became a member of the African National Congress (ACN), which campaigned for the right of the black inhabitants of the colony of South Africa to land ownership and full status. The Indian National Congress and its methods of non-violent resistance, espoused by M. Gandhi, served as their model.


6. In the early 1970s, the Soviet Air Force expressed a desire to replace the MiG-21 with a more modern twin-engine fighter aircraft with greater air combat capabilities. Yakovlev submitted a series of designs under the designation Yak-45.


7. Soldier of the Austro-Hungarian monarchy, Serbia, Russia and Czechoslovakia. Legionnaire, general and prisoner and victim of the Gestapo.


8. Hiero II was a Greek tyrant of Syracuse between 275 and 215 BC and the illegitimate son of the Syracusan nobleman Hierocles, who claimed the house of Gelon. He was a former general of Pyrrhus of Epirus and an important figure in the First Punic War. He figures in the story of the famous thinker Archimedes, who exclaimed "Heureka".


9. David Ben-Gurion was Israel's first prime minister; he held the post on and off for thirteen years. Ben Gurion's enthusiasm for Zionism culminated in his role in the founding of Israel. After leading it to victory during the War of Independence, he helped to establish state institutions and oversaw the absorption of hundreds of thousands of Jews from around the world.

each annotation above corresponds to a number and placement in two-dimensional space. The points closer together form something like a "cluster of similar texts", notice for example the grouping of 1 and 8 in the graph on the right, i.e. the articles about personalities from before our time. Next to it are 5-7-9, also personalities, but more from the recent past. And on the other side weapons, 3 and 2 - armoured vehicles, and a little further 6, an aircraft and 4 a tank ... it is only a small sample of data, but I think it is obvious that even on Czech and Slovak texts it is quite solid to determine the "similarity" and on the basis of this I could recommend other articles similar at the end of each article.




The second area that came to my mind while researching the first one is the creation of an automatic "annotation" of the article, i.e. summarizing what the article is about. Again, this relies on the AI's ability to read the text and, according to some specification (short, medium, long, in sentences or paragraphs), write out a summary. The interesting thing is that it is able to absorb and understand Czech and Slovak text, but the answer is only in English, which is not a problem, I just need to run DeepL through the translator and deliver the Czech version of the same. For the sake of interest, I am again attaching three "summaries" that it created based on the text of our three relatively recent articles:











it would be useful in two places, firstly when writing the article, so that one does not have to come up with an annotation, and secondly in the place where one nominates a topic here from the forum as an article, and one has to insert the NAME of the article and the ANNOTATION. And not many people do not do those nominations, I think it's because of the fact that to come up with that one or two sentences are actually a terrible problem In the new version of valka.cz I don't show the annotations on the front page of the article, but it's more of an experiment, if it's better or worse without them, they can come back at any time, so I definitely want to keep this field next to the article. The way it would work is that when nominating or writing an article, the AI would "read" our text and add a summary, which the human could of course tweak as needed, but it wouldn't have to write it all from the beginning.



I'm interested in your opinion, do you find this way useful, is "Skynet" in this state capable enough to be useful to us, or do you think we won't use it yet?

That's the question - just with annotations.
Yes, AI saves effort and can pull out everything essential from the article in a short paragraph
No, it clashes with your requirement to have text in the annotation that catches the eye and inspires to open the article
But why not try it?

I'm all for it, at least it's worth a shot. Both for annotations and for offering similar articles.


I'd like to see something that goes through our extensive database and pulls out "candidates" for articles even if they don't have a percentage set, or if it's unnecessarily low. "Hand" crawling is very tedious...

Janko : it probably can't do that yet, but we can look at it differently - we usually nominate things for an article that have (besides being interesting, of course) a minimum length. So I can try to make some kind of a tool that looks which topics are "full of text", where there are more than N characters or words or something like that, and where nothing is nominated. Exclude discussion threads and posts, and see if that gives some interesting list. Which of course will have to be scrolled through by hand, but at least the choice of where to look will be narrowed down a bit.


That it can't write "engaging" text is of course a fact, but I wonder if we have that many creatives among us who would take the plunge. As far as the articles are concerned, I think that Janko and Rad have a lot on their shoulders, and it's more on the organizational or linguistic side. Unfortunately, there are as many of us as there are, and I don't know if we will be able to find someone else who could work in this area. And where there are no hands, the brain has to step in

Už dlhšie sa hrám s Bardom a skúšal som ho zneužiť pri tvorbe nejakých vecí na valku. Z hľadiska všeobecnej histórie je ako tak použiteľný, detaily mu nejdú (množstvo informácií je v klasickej knižnej podobe, ku ktorým nemal a nemá prístup a aj množstvo webových zdrojov je pre neho nedostupných). Takže AI používam najmä na tvorbu popisov k videu a učesanie prekladov - popisov k svojim videám na youtube + na to, aby môjho synátora učil matiku (o pár mesiacov maturuje a je z nej troška mimo).

Well AI is cool, but when something comes out of itself so it clearly still has to be somehow cut off so that it's not... machine-like. And when it's done nicely it can give the feeling that it's also creative