i had so many discussions about ai those past 3 or 4 years, some were good, some were bad, some were interesting, most of them were either very frustrating or very anger-inducing. and sometime, during those conversations, happen something very common : someone says something, and i don’t have a rebuttal, or someone interrupt the conversations and we move on to another subject. or sometime it’s just during a job interview and i prefer not to offer a rebuttal and just nurse my anger in a deep part of my brain. and after a while, often in the shower, i replay the argument in my head, and find new ways to respond, new answers to give to my now imaginary opponent. and it’s frustrating that only my bottle of shampoo will ear those arguments, so i decided, hey what the hell, that’s why this blog exist : to give an out to my most annoying thoughts and shower the world with them.

so i’m compiling them into an article.

if new discussions emerges, i’ll be posting more.

(but not updating this one that would be silly i want more articles)

🇫🇷 il n’y a pas de version en français pour l’instant, mais je verrais si je me sens d’en faire une traduction. en attendant, vous pouvez utiliser la fonction “traduire cette page” de votre navigateur internet.

ai will allow me to live my life and stop working

no it’s not.

well, ai itself would maybe help with that if you lived in a socialist utopia. after all, an utopia is simply a slave state where you manages to hide the slaves and as unconscious automatons, ai make for the perfect entities to take care of essential work. but first of all we’re not there yet and second of all we would have to calculate if the energy and resource intensive time to train ais on all those different job is really worth it and compatible with what we have left as finished resources on our planet. maybe over a very long period of time to not overstress our energy production with unnecessary model training?

but that’s in a perfect world; in our reality, nobody and i mean NOBODY plans to give you a salary to do nothing while an ai is doing your job. what is planned is to get you fired and replaced with an ai that will do your work but cost less to your boss that will be replaced by an ai the second their company is bought by a bigger one and so on until the only companies left are ai companies selling to themselves models and so on, you get it, capitalism at it again. and you, during that process, will be left to die, or do the few manual job ai will not crack because after all LLM are not a 1:1 equivalence with the human brain and some things will always be above their capacities, even if we manage to put them in bodies better than our owns.

ai makes my homework easier, i don’t have to read the book, chatgtp write my paper for it

this is a discussion i had with my nephew : and yes, this is very true, all of this.

but it’s not good. those things are not supposed to be easy, and they are not made to be interesting. they are the annoying cardio exercises your coach makes you do before training : when you are not making the effort of reading and then writing something, whether it is a book review, a commentary, a text analysis, what is expected of you is not to fill up a piece of paper because it’s good for the economy that you use ink somehow. you’re ask of doing this because learning and practicing those skills is something incredibly important and valuable in a democratic society. what you are learning during those steps is to not only read a text but understand it, internalize it, criticize it, and help you shape what you think and who you are as a person.

using a machine to do that for you, means that you are delegating your own understanding of reality to someone else. and i didn’t say “something else”, i said someone else, because elon musk and grok proved it : ai is first and foremost owned by someone, and therefore, the projection of the belief system of that person. so, if you rely on that person, you are denying yourself your own independence. television and cable channel did that for years but ai, by the place it’s taking in your life, is doing so much worst, and quicker.

and once those skills are gone, they are hard to get back, because they require an conscious effort. a conscious effort that you wont want to make because the machine will convince you that you don’t need to. and then, the machine, or rather the people exploiting the machine, will be able to sell you anything and everything. nobody is immune to propaganda, but YOU, you will become a sponge to it.

and that count for kids in school, but also for you who use ai to summarize meetings and emails. you don’t realize it, but your brain and your critical sense is being melted away.

ai doen’t consume too much when you put what it brings in perspective of what it can do

often this argument is brought up as “when ai will fix global warming”. and i want to say to that : where? where is ai fixing global warming? where is ai fixing its own energy and resources consumption? for now, all i can see are some very very specific ai specifically trained for fundamental research project, medical purposes (not chatgtp telling you you have cancer, but rather ai trained on xray to detect cancer more accurately), some old models are still improved in some weather research centers…

but if we are building bigger and bigger datacenters today, it’s not to bring power to those models, it’s to bring power to the ones made to replace your cousin working in a call-center, make your saas startup generate even more technical debt via claude or generate naked pictures of children. if elon musk is installing gaz powered electrical power station, it’s not to improve how to calculate the oxygen needed in an hydroponic farm of the future, it’s to make sure grok doesn’t make the mistake of considering trans people as real people.

the goal of the ai industry is not to save the world, it’s to save their own skin by making the economy dependent from them, and nothing else.

all datacenters consume water and energy, we didn’t see you complain before

so a weird part of my career is that i started as an adminsys so i have a beginning of a response for you that is a bit technical but i’m going to try and dumb it down. basically it comes from the difference between an AI datacenter and a standard, non-specific datacenter : very basically, AI datacenters run at maximum, all the time. they also consume energy in a very different way. we spent years working on virtualizing systems, i.e. trying to fit as many systems as possible on one single CPU, on one single core of a CPU (so a fraction of it) specifically to save running cost. so we created infrastructure where we shared everything; we shared ram, we shared storage, we shared computation capacity. this way, datacenters stayed relatively ok-sized (small compared to ai datacenters) to the point that, you might not know, but if you live in a city, you might live closer to an old school datacenter without knowing it. the one i used to intervene on was just a weird hidden warehouse smack down in marseille third district, in the middle of hotels, hospitals and residential buildings. and we still need those, they are as capital to our modern life infrastructure as the power-grid, train rails or any supply chain, because they provide our current internet, it provides services that we use daily to pay for stuff, know where we are, shop, order food, watch dumb cat video and work.

also, GPUs and CPUs didn’t followed the same development path AT ALL : if you look at CPUs those last few decade, the goal was to increase calculation, increase transistors concentration, while keeping power consumption at bay, because power consumption meant heat, heat meant low performances and potential wear. GPU does : none of that. the nvidia way those last few years was “just buy bigger power supplies”, and once ai computation came around and they became the main focus of the economy, they didn’t batted an eye and kept pushing.

so yeah, regular datacenters gobble energy and water too, but it’s nothing compared to ai datacenters AND they are doing an important job in our modern infrastructure and life. ai datacenters do not, i don’t care how important you feel is your ai girlfriend, your vibecoding buddy or your ghibli picture filter, but they are NOT as important.

the ai genie is out of the bottle now, we HAVE to use it, we can’t go back

the fuck we can’t. of course we can. there is in fact a precedent : asbestos.

asbestos, when introduced industrially and mostly in construction, was a miracle technology. it was used everywhere, praised everywhere. but as we discovered that it was highly carcinogenic, we decided, as a society,that its benefits were too few to keep it around, and labelled it culturally as well as literally as something very toxic, almost synonymous as nuclear waste. the only things convincing you that you HAVE to keep ai around is because it made you very lazy, gave you undeserved and untrained skills, and because like people in the 60s putting lead fuel in their car, or construction workers putting asbestos in the celling of a house, you are meant to not understand the consequences of what you are doing and just reap the fake “benefits”, while the ones pushing that the ai revolution is inevitable, are the ones reaping the real benefits.

yeah, that’s the last point; ask yourself who’s making you say that? is it you? your own laziness? or is it someone really rich that see in your addiction a way to get them even richer?

ai must be consuming as much as a video game studio, it must be as costly to make a game like gta6

yes this is a real phrase that has been spoken to me by a CEO during an interview

i don’t really know how to respond to this one, except by saying a big fat “no”, like, yes making a game and playing a game consume a lot of energy but so is playing a game on your phone or even playing cards with a friends. but i never heard of a video game studio that had to open a gas-fired power plant inside their workspace to allow them to make a game.

of all the arguments used in front of me to defend ai, i think that one was the dumbest. by a large margin.

ai doesn’t consume that much energy, we need more studies

same dude as above btw

this is just a lie. it’s not even a clever lie or a smart lie it’s just a dumb lie : the ai industry consume today more than 3% of the energy consumed by all humans on earth and that slice of the energetic pie will be, by their own account, growing a lot in the next 4 years.

ai will be 100% right at some point

not it will not, for multiple reasons :

  • ai improvement are made by feeding it increasingly bigger and bigger dataset, but those datasets will contains errors, which will always ALWAYS create hallucinations.
  • ai has no way to know it’s hallucinating. it’s basically a big fat autocorrect system that is just statistically guessing what’s supposed to be its next answer; you can increase the usage by making it check its results but that’s just using ai over ai, that doesn’t solve the main issues.
  • ai is starting to running out of “human content” and is starting to use ai content for its training, which is only making hallucinations worst.
  • in term of raw performances, it’s already plateauing, what’s improving are the tools around ai and those will reach their peak at some point and like very product, will start going downward in quality.

ai will increase productivity

except i personally don’t believe it’s always true. when we talk about gains i look at 2 things :

  • what are we talking about? is it gain in production of code? is it gain in features? in stability? in refactoring? in quality?
  • who is talking about those gains? is it a research institute? is it a company heavily using and emphasizing their internal ai deployment in their own teams? is it someone at anthropic or open ai?

for the first one, and i talk of experience, most often what is sold as a gain in productivity is the fact that you can ship a feature faster. which is very much true, ai made it so easy to ship something. but ask any dev : shipping was never the hard part. actually, making a feature, any feature, in a void, is very easy. you know the classic “they are asking me 2 weeks to add a button” which is supposed to show how hard it is to do simple tasks without breaking things, well ai now empower people to do that change and to not worry about the consequences of that change. will it break something? you don’t know, you’re not the one changing the button anymore. have you took the time to review what the ai produced? of course not, it would be slowing down the push to shipping. with ai, you don’t have ownership of what you’re doing, you just herding automated processes toward something you want and you cross every fingers on your body that token price will not hike more and that your basecode is still somewhat readable.

because chances are : they are trying to convince you to buy something, or just are trying to reasure their shareholder that yes, that 80k a month in tokens was worth it.

ai made code easy

no ai made production easier, it always was the easy part, but it made maintenance worst than before. and it killed the capacity for junior to learn how to properly code, which was, again a very good way for them to ensure selling their product even after prices explode, because you have now created an economy where people completely forgot how to work without your product. and government are encouraging this, how f u n.

also, like, what the actual fuck? we spent the last 3 decades at least yelling that making black boxes is a really bad thing, that obfuscating code was a big no-no. and now we created a machine that only make that? code that nobody knows how it works? that nobody review? that nobody knows how to fix?

I N S A N E

you are just jealous because ai is smarter than you

no, i know a machine that is supposed to know everything look smarter than me, but my understanding of my absence of universal knowledge is what makes me, in the end, infinitely smarter than it is.

the fact that you consider yourself dumber than an ai, that, is the concerning part.

there, im gonna stop here. concerning sources of everything i talk about in this article, you’re grownups, i can’t hold your hand about everything (and i never see those fucking bring up sources first so i don’t know why i should do be held to a higher standard).

i still have more points to talk about, more development to do, but that will have to wait for a future part article, as i think this will be a recurring series on the blog, and also i have to stop at some point and click the publish button or that article, like my eternal frustration about this subject, will never end.