M contents/articles/gen-ai.md → contents/contents/articles/ai.md

@@ -1,5 +1,5 @@ 
 -----
-id: gen-ai
+id: ai
 draft: true
 title: "A Tale of Hype and Fallacy"
 subtitle: "Reflections on Generative AI, sparkles, and... washing machines"
@@ -7,21 +7,23 @@ content-type: article
 timestamp: 1737926162
 -----
 
-I don't write very often these days. But if and when I do though, it should be about something meaningful, so than in [ten or twenty years](/articles/twenty-years/) time I will be able to say "Oh wow, that thing really felt important back then, and it isn't now" or "I didn't really understand the magnitude of that at the time".
+I don't write very often these days. But if and when I do though, it should be about something meaningful, so than in ten [twenty years](/articles/twenty-years/) time I will be able to say "Oh wow, that thing really felt important back then, and it isn't now" or "I didn't really understand the magnitude of that at the time".
 
-That's why I have decided that I should really write about Artificial Intelligence. Not the _real thing_ (AGI) that we will probably have twenty years from now, just the dumb-but-sometimes-useful surrogate that is _generative_ AI, that we have today, in 2025.
+That's why I have decided that I should really write about Artificial Intelligence. Not the _real thing_ that is still science fiction, just the dumb-and-deceiving, slop-generating surrogate that is _generative_ AI, that we have today, in 2026.
+
+> %notice%
+> **Note** 
+> 
+> I started writing this article about a year ago, and didn't finish it because I lost interest in it. After a year, I tried to crystallize my thoughts now that I have a clearer understanding and therefore clearer stance on the topic.
+> 
+> TL;DR: generative AI is probably the worst piece of technology ever invented.
 
 ### Not your average positronic brain
 
 Back in 1987 (nearly FORTY years ago, GOD I am old!), when Star Trek: The Next Generation premiered, they introduced the character of Lt. Cmdr. Data, a sentient android, portrayed by the legendary Brent Spiner. That character was probably thought originally as a sort of replacement for Mr. Spock from TOS, being primarily driven by logic, but with a _twist_: even though it couldn't understand basic jokes, or even speak using word contractions, Data longed to become human, above all else. 
 And his positronic brain, in all its sophistication, was not able to feel emotions (he eventually got an upgrade, but that's another story). Not only that, he was struggling to even _simulate_ them, and was incapable of lying.
 
-Today's LLMs are nothing like Data. Our poor man's AI is a powerful set of algorithms that not only is able to use contractions and even effectively mimic the writing style of humans, it is _frequently_ wrong, it hallucinates, and it has a natural talent to deceive humans. It can also effectively simulate creativity, even paint you a brand new picture in less than a second!
-
-<figure>
-	<img src="/images/ai/ai-world-domination.webp" alt="An AI-generated image of AI taking over the world" />
-	<figcaption>Create an image representing Artificial Intelligence conquering the world and the internet. Overemphasize the technology side of it and draw your inspiration from popular science fiction.</figcaption>
-</figure>
+Today's LLMs are nothing like Data. Our poor man's AI is a powerful set of algorithms that not only is able to use contractions and even effectively mimic the writing style of humans, it is _frequently_ wrong, it hallucinates, and it has a natural talent to deceive humans. It can also effectively simulate creativity, even paint you a brand new picture in less than a second. 
 
 What about quality, you ask? Ehhh we are working on it. What about reliability, can it actually... Pffft, nevermind. 
 
@@ -33,11 +35,11 @@ TRUST
 
 (generative) AI
 
-No really, just don't. Tell your children, your grandmothers, all your loved ones that Chat-thingie or whatever crap it's called should not be trusted more than a used car salesman, _at best_ (it's a figure of speech, I actually met a few pretty decent used-car salesmen actually).
+No really, just don't. Tell your children, your grandmothers, all your loved ones that Chat-thingie or whatever crap it's called should not be trusted more than a used car salesman, _at best_ (it's a figure of speech, I met a few pretty decent used-car salesmen actually).
 
-### It is actually pretty good (at certain things)
+### It kinda works on certain things
 
-Assuming you are not dumb enough to trust them with your life, your job, or anything important, LLMs are undeniably one of the greatest inventions of the decade. Despite that it changed forever our imaginary when it comes to AI and killed 90% of science fiction literature with it, these things are pretty damn impressive, for certain things.
+Assuming you are not dumb enough to trust them with your life, your job, or anything important, LLMs could actually be useful at certain things. Despite that it changed forever our imaginary when it comes to AI and killed 90% of science fiction literature with it, these things are pretty damn impressive, for certain things.
 
 Things like:
 - Understanding written text
@@ -54,7 +56,10 @@ > "LLMs are good at transforming text into less text"
 
 The above quote should be turned into mobile and desktop wallpapers, printed on billboards, and tattooed on the forehead of execs.
 
-Perhaps it can be a bit of an oversimplification, but again, it doesn't say that they are "only" good at that, right? I think it's a good and safe rule of thumb, for when you are in doubt.
+Again, your mileage may vary. The worst thing about it? Non-determinism. It gets something right today, won't get it right tomorrow, or at least, you don't have the mathematical certainty about it. 
+
+And well, this sucks.
+
 
 ### Tutti mi vogliono, tutti mi cercano
 
@@ -94,6 +99,8 @@ You really cannot sell generative AI by itself: you need a problem to solve, and you need to solve it in a way that actually makes things easier or faster for the user. AI _can_ be a productivity boost, but it cannot replace the product itself.
 
 Let's go through a couple of examples. 
 
+This is from last year. Eons in AI terms, but still worth of a mention.
+
 As part of a mini hackathon, one of the engineers desperately wanted to showcase how they managed to integrate AI into documentation search. I was expecting the usual _here's a summary of the results in a blurb_ kind of thing, but it wasn't even that. Here goes:
 
 Engineer: _"So, I am integrating AI into our search command. See, from the command line I am doing a search on our documentation,I fetch the first three results, and then I send them to AI and..."_
@@ -106,7 +113,8 @@ Me: _"So you are saying that AI is actully _changing the page title_ and providing a sort of preview of the content. But... would the user really want that, or would the user just be happy with seeing the _same_ page title and the search-highlighting provided natively by the existing fulltext search?"_
 
 Next...
 
-Another example was a pseudo Copilot thing that was able to process a prompt asking to add an icon indicator to a specific object type. As always, I was given the usual demo, and the results were pretty good.
+Another example (again from
+The same time) was a pseudo Copilot thing that was able to process a prompt asking to add an icon indicator to a specific object type. As always, I was given the usual demo, and the results were pretty good.
 
 Me: _"OK, can I then use this copilot thingie to implement other customizations? Like for example [...]"_
 
@@ -137,7 +145,7 @@ > In our own day, the Church offers to everyone the treasury of her social teaching in response to another industrial revolution and to **developments in the field of artificial intelligence** that pose new challenges for the defence of human dignity, justice and labour.
 
 -- [Pope Leo XIV names AI one of the reasons for his papal name, The Verge](https://www.theverge.com/news/664719/pope-leo-xiv-artificial-intelligence-concerns)
 
-That is big news. It's not that the Catholic Church is going to excommunicate people for using ChatGPT because it's the eevil or something, but it is going to probably spend a significant amount of time helping people how to deal with AI. I find the fact that it is being compared to the industrial revolution as incredibly fitting. I am willing to bet that our new Pope will be writing an encyclical about the role of generative AI in our sociaty, like his namesake did at the time of the industrial revolution.
+That is big news. It's not that the Catholic Church is going to excommunicate people for using ChatGPT because it's the evil or something, but it is going to probably spend a significant amount of time helping people how to deal with AI. I find the fact that it is being compared to the industrial revolution as incredibly fitting. I am willing to bet that our new Pope will be writing an encyclical about the role of generative AI in our sociaty, like his namesake did at the time of the industrial revolution.
 
 While we wait, there's actually a very interesting (and very, VERY long) essay written by... the Dicastery for the Doctrine of the Faith and the Dicastery for Culture and Education that already provides a rather in-depth analysis of artificial intelligence and its present and future impact on society: [Note on the Relationship Between Artificial Intelligence and Human Intelligence](https://www.vatican.va/roman_curia/congregations/cfaith/documents/rc_ddf_doc_20250128_antiqua-et-nova_en.html).