feat(blog): are the LLMs worth it

2 files changed, 226 insertions, 1 deletions

diff --git a/content-org/blog.org b/content-org/blog.org
index 39fa134..39a4e70 100644
--- a/content-org/blog.org
+++ b/content-org/blog.org
@@ -7,7 +7,7 @@
 #+HUGO_WEIGHT: auto
 #+HUGO_SECTION: blog
 
-* 2024  [43/44]                                                       :@blog:
+* 2024  [44/45]                                                       :@blog:
 :PROPERTIES:
 :EXPORT_HUGO_SECTION: blog/2024
 :EXPORT_HUGO_CUSTOM_FRONT_MATTER+: :image_dir "blog/images" :image_max_width 600
@@ -64,6 +64,118 @@ There will be no official XMPP bridge anytime soon, and therefore there will be
 What we need is a great product
 /We've got the technology/.
 
+
+** DONE Are the LLMs worth it?
+CLOSED: [2024-06-30 Sun 20:06]
+:PROPERTIES:
+:EXPORT_FILE_NAME: are-llms-worth-it
+:EXPORT_HUGO_CUSTOM_FRONT_MATTER: :abstract The only way to fight I see
+:END:
+
+Recently, half of my waking time seems to be spent actively ignoring the current AI boom.
+Every investor, every product, even some open source project - all are part of the biggest craze I've ever witnessed.
+My skepticism for all this /crap/ doesn't come from being against the idea behind it.
+I love AI, but LLMs are different.
+
+How do you measure if something is worth it?
+I guess it's something akin to a simple equation:
+
+
+#+begin_src shell
+  c = v * N
+#+end_src 
+
+Where
+c - cost
+v - value
+N - your personal acceptance factor
+
+So, a product is worth the cost, if it gives enough value based on how much you are willing accept.
+What is the cost?
+It's hard to tell.
+What is value?
+It's even harder to tell.
+
+When it comes to LLMs, the cost is /huge/.
+It may very much be the most costly tech we will see in our lives.
+It costs very little when we limit ourselves to /monetary value/.
+It's a few bucks here, a few bucks there.
+Anyone into subscriptions most likely already has bigger sinkholes.
+
+But that approach is significantly shortsighted; reduced to minimum.
+The actual cost of the technology is far more reaching.
+
+Firstly, the development.
+VCs all around the globe can't throw their money fast enough.
+It seems that just adding "AI" to a shampoo gives significant chance of a hefty A series.
+So much of the current world economy is based on the idea that one day all that investment will be paid.
+Just look at the stock value of Nvidia!
+It seems that just yesterday they were known primarily for overpriced graphic card for gamers.
+But they were lucky, their product powered the last two big crazes - first crypto, now LLMs.
+
+The second aspect of this part of life cycle is the man power.
+Just how many engineers are now tasked with integrating every, imaginable system with ChatGPT?
+And how many product managers are forced to think of an idea of what that integration is supposed to do?
+Those product are, almost aways, /worse/ in the process.
+Even something as simple as Expensify became unusable when they changed their product into a chatbot.
+
+And then comes the running cost that will be paid by our children.
+The environmental impact is unimaginable.
+We're producing huge amounts of air pollution, using water and rare minerals like there's no tomorrow.
+For the last century or so, we've abused our planet, but then we've decided that enough is enough.
+Time to think of survival and become /green/.
+All this is thrown out the window.
+No survival of human race will stop Altman from becoming a multi billioner!
+
+Then there are the social costs.
+We're just starting to see them, but it's already dreadfully.
+LLM companies are treating all human creation as a free buffet, ignoring everything we've had before.
+The web, as we grew to know and love, may soon come to an end.
+Sometimes even I have hard time justifying existence of this site.
+I have a lot of fun, do something here almost daily.
+But in reality I make the /assholes/ richer, as they will harvest everything here and put in their huge data models.
+
+And, of course, people are starting to lose their jobs.
+Not that many yet, as the CEOs need to be crazy enough, but it's starting.
+And it will only get worse, as more and more product /promise/ automation.
+Most likely it will happen before the promise is fulfilled, but that's beside the point.
+
+The costs are huge, but what are the benefits?
+I've tried to read what all the proponents say, and the responses I've found differ very slightly:
+some small auto generation, here and there.
+Nothing grand, a few dozen minutes of manual labor each time - unless we're talking about image/multimedia generation.
+You ma even get what you want it to be, assuming you've already got some experience in prompting.
+Mind you, when an LLM works, it works like magic.
+It is, however, not the case most of the time.
+You get a result, you try to convince the machine that there is something wrong, you get a result, you find something and so on.
+Up until it's ready for manual fixes.
+
+Looking at this, we've got a huge =c=, very small =v=.
+If you say that this is worth it, your =N= needs to be huge.
+
+But there's this notion, that we're just at the start.
+We've got the first version and there will be huge breakthroughs soon.
+Either the cost needs to be decreased, or the value needs to skyrocket.
+Both are possible.
+There's enough money thrown for there to be more people working on those problems than there are scientists working on curing cancer.
+Will they achieve anything of value?
+We can /guess/, but guessing will not change the fact, that the current [[https://link.springer.com/article/10.1007/s10676-024-09775-5][bullshit machines]] are *not* worth it.
+
+But let's be realistic here: betting on /scientific breakthrough/ is not something a sane mind should do.
+Especially not one data-driven!
+Events like this rarely happen, and the unravel ChatGPT 2 proved that we've already witnessed one, just a few short years ago.
+The next one may take years, or decades.
+We will be paying the cost until the LLM craze ends, and perhaps it will not even destroy us before that.
+
+Personally, I am sure that AI Winter is inevitable.
+And it will be winter like we've never seen before, as never before so much money was thrown into furnace.
+There are /great/ uses of AI - just think of image recognition, or weather forecasting.
+All of this may be lost, as one fucking Altman stole all the air from the room, just to burn us all in the process.
+I am afraid that the winter will not only stop significant work on LLMs, but for general Artificial Intelligence.
+
+Note, that I ignore all the /risks/ here, and just look a the technology.
+But the costs are even higher, as we should put things which it enables - disinformation for one.
+
 ** DONE LLM honeypot
 CLOSED: [2024-06-28 Fri 14:14]
 :PROPERTIES:
@@ -3506,3 +3618,4 @@ https://www.youtube.com/watch?v=kwz-Md6OoyA
 
 
 
+
diff --git a/content/blog/2024/are-llms-worth-it.md b/content/blog/2024/are-llms-worth-it.md
new file mode 100644
index 0000000..53c9022
--- /dev/null
+++ b/content/blog/2024/are-llms-worth-it.md
@@ -0,0 +1,112 @@
++++
+title = "Are the LLMs worth it?"
+author = ["Michał Sapka"]
+date = 2024-06-30T20:06:00+02:00
+categories = ["blog"]
+draft = false
+weight = 2002
+abstract = "The only way to fight I see"
++++
+
+Recently, half of my waking time seems to be spent actively ignoring the current AI boom.
+Every investor, every product, even some open source project - all are part of the biggest craze I've ever witnessed.
+My skepticism for all this _crap_ doesn't come from being against the idea behind it.
+I love AI, but LLMs are different.
+
+How do you measure if something is worth it?
+I guess it's something akin to a simple equation:
+
+```shell
+c = v * N
+```
+
+Where
+c - cost
+v - value
+N - your personal acceptance factor
+
+So, a product is worth the cost, if it gives enough value based on how much you are willing accept.
+What is the cost?
+It's hard to tell.
+What is value?
+It's even harder to tell.
+
+When it comes to LLMs, the cost is _huge_.
+It may very much be the most costly tech we will see in our lives.
+It costs very little when we limit ourselves to _monetary value_.
+It's a few bucks here, a few bucks there.
+Anyone into subscriptions most likely already has bigger sinkholes.
+
+But that approach is significantly shortsighted; reduced to minimum.
+The actual cost of the technology is far more reaching.
+
+Firstly, the development.
+VCs all around the globe can't throw their money fast enough.
+It seems that just adding "AI" to a shampoo gives significant chance of a hefty A series.
+So much of the current world economy is based on the idea that one day all that investment will be paid.
+Just look at the stock value of Nvidia!
+It seems that just yesterday they were known primarily for overpriced graphic card for gamers.
+But they were lucky, their product powered the last two big crazes - first crypto, now LLMs.
+
+The second aspect of this part of life cycle is the man power.
+Just how many engineers are now tasked with integrating every, imaginable system with ChatGPT?
+And how many product managers are forced to think of an idea of what that integration is supposed to do?
+Those product are, almost aways, _worse_ in the process.
+Even something as simple as Expensify became unusable when they changed their product into a chatbot.
+
+And then comes the running cost that will be paid by our children.
+The environmental impact is unimaginable.
+We're producing huge amounts of air pollution, using water and rare minerals like there's no tomorrow.
+For the last century or so, we've abused our planet, but then we've decided that enough is enough.
+Time to think of survival and become _green_.
+All this is thrown out the window.
+No survival of human race will stop Altman from becoming a multi billioner!
+
+Then there are the social costs.
+We're just starting to see them, but it's already dreadfully.
+LLM companies are treating all human creation as a free buffet, ignoring everything we've had before.
+The web, as we grew to know and love, may soon come to an end.
+Sometimes even I have hard time justifying existence of this site.
+I have a lot of fun, do something here almost daily.
+But in reality I make the _assholes_ richer, as they will harvest everything here and put in their huge data models.
+
+And, of course, people are starting to lose their jobs.
+Not that many yet, as the CEOs need to be crazy enough, but it's starting.
+And it will only get worse, as more and more product _promise_ automation.
+Most likely it will happen before the promise is fulfilled, but that's beside the point.
+
+The costs are huge, but what are the benefits?
+I've tried to read what all the proponents say, and the responses I've found differ very slightly:
+some small auto generation, here and there.
+Nothing grand, a few dozen minutes of manual labor each time - unless we're talking about image/multimedia generation.
+You ma even get what you want it to be, assuming you've already got some experience in prompting.
+Mind you, when an LLM works, it works like magic.
+It is, however, not the case most of the time.
+You get a result, you try to convince the machine that there is something wrong, you get a result, you find something and so on.
+Up until it's ready for manual fixes.
+
+Looking at this, we've got a huge `c`, very small `v`.
+If you say that this is worth it, your `N` needs to be huge.
+
+But there's this notion, that we're just at the start.
+We've got the first version and there will be huge breakthroughs soon.
+Either the cost needs to be decreased, or the value needs to skyrocket.
+Both are possible.
+There's enough money thrown for there to be more people working on those problems than there are scientists working on curing cancer.
+Will they achieve anything of value?
+We can _guess_, but guessing will not change the fact, that the current [bullshit machines](https://link.springer.com/article/10.1007/s10676-024-09775-5) are **not** worth it.
+
+But let's be realistic here: betting on _scientific breakthrough_ is not something a sane mind should do.
+Especially not one data-driven!
+Events like this rarely happen, and the unravel ChatGPT 2 proved that we've already witnessed one, just a few short years ago.
+The next one may take years, or decades.
+We will be paying the cost until the LLM craze ends, and perhaps it will not even destroy us before that.
+
+Personally, I am sure that AI Winter is inevitable.
+And it will be winter like we've never seen before, as never before so much money was thrown into furnace.
+There are _great_ uses of AI - just think of image recognition, or weather forecasting.
+All of this may be lost, as one fucking Altman stole all the air from the room, just to burn us all in the process.
+I am afraid that the winter will not only stop significant work on LLMs, but for general Artificial Intelligence.
+
+Note, that I ignore all the _risks_ here, and just look a the technology.
+But the costs are even higher, as we should put things which it enables - disinformation for one.