I want to let people know why I’m strictly against using AI in everything I do without sounding like an ‘AI vegan’, especially in front of those who are genuinely ready to listen and follow the same.
Any sources I try to find to cite regarding my viewpoint are either mild enough to be considered AI generated themselves or filled with extremist views of the author. I want to explain the situation in an objective manner that is simple to understand and also alarming enough for them to take action.
at work mgmt always brings it up. “we need to use it more!”.
I say nothing. I smile and nod. I ignore AI prompts. I ignore emails written by AI. I ignore requests coming in to integrate AI into the product.
nobody has asked me about any of the inaction for the last year so I don’t plan on drawing any attention to it by outing myself.
edit: I suppose if anybody does I can just say the AI agent I used failed to alert me to the thing they wanted. 🤣
Don’t bring it up unless prompted or posing philosophical questions to family and friends. I once was creating a video for a client that sent me some generated images into the video that he thought were hilarious. I told him sorry, no, didn’t over explain and just said that he would need to hire somebody else if he wants those things. It’s not very hard. I always try to push for human artists and explain not only is it better for the ecosystem but for copyright reasons.
I just mentioned to a friend of mine why I don’t use AI. My hatred towards AI strives from people making it seem sentient, the companies business model, and of course, privacy.
First off, to clear any misconception, AI is not a sentient being, it does not know how to critical think, and it’s incapable of creating thoughts outside from the data it’s trained on. Technically speaking, a LLM is a lossy compression model, which means it takes what is effectively petabytes of information and compresses it down to a sheer 40Gb. When it gets uncompressed it doesnt uncompress the entire petabytes of information it uncompresses the response that it was trained from.
There are several issues I can think of that makes the LLM do poorly at it’s job. remember LLM’s are trained exclusively on the internet, as large as the internet is, it doesn’t have everything, your codebase of a skiplist implementation is probably not going to be the same from on the internet. Assuming you have a logic error in your skiplist implementation, and you ask chatGPT “whats the issue with my codebase” it will notice the code you provided isn’t what it was trained on and will actively try to fix it digging you into a deeper rabbit hole then when you began the implementation.
On the other hand, if you ask chatGPT to derive a truth table given the following sum of minterms, it will not ever be correct unless heavily documented (IE: truth table of an adder/subtractor). This is the simplest example I could give where these LLMs cannot critical think, cannot recognize pattrrns, and only regurgitate the information it has been trained on. It will try to produce a solution but it will always fail.
This leads me to my first point why I refuse to use LLMs, it unintentionally fabricates a lot of the information and treat it as if it’s true. When I started using chatGPT to fix my codebases or to do this problem, it induced a lot of doubt in my knowledge and intelligence that I gathered these past years in college.
The second reason why I don’t like LLMs are the business models of these companies. To reiterate, these tech billionaires make this bubble of delusions and fearmongering to get their userbase to stay. Titles like “chatGPT-5 is terrifying” or “openAI has fired 70,000 employees over AI improvements” they can do this because people see the title, reinvesting more money into the company and because employees heads are up these tech giants asses will of course work with openAI. It is a fucking money making loophole for these giants because of how many employees are fucking far up their employers asses. If I end up getting a job at openAI and accept it, I want my family to put me into a god damn psych ward, that’s how much I frown on these unethical practices.
I often joke about this to people who don’t believe this to be the case, but is becoming more and more a valid point to this fucked up mess: if AI companies say they’ve fired X amount of employees for “AI improvements” why has this not been adopted by defense companies/contractors or other professions in industry. Its a rhetorical question, but it makes them conclude on a better trajectory than “the reason X amount of employees were fired was because of AI improvement”
If you were wondering if chatGPT can do a sum of minterms here’s one I derived on the fly, I’ve attached screenshots of the conversation we had…
There isn’t a way to use AI in good faith.
Either you are ignorant of the tech and its negative effects, or you arent.
Depending on how hardcore you are about it, you can’t.
Are you getting up in people’s face to tell them not to use it, or are you answering why you choose not to use it?
Are you extremely strict in your adherence? Or are you more forgiving based on the application or user?There are two general points I like to make:
- Big companies are using it to steal the work of the powerless, en masse. It is making copyright strictly the tool of the powerful to use against the powerless.
- If these companies aren’t lying and will actually deliver what they say they’re going to deliver in the timeline they stated, then it’s going to cause mass unemployment, because even if (IF) this creates new jobs for every job it destroys, the market can’t move fast enough to invent these new careers in the timeline described. So either they’re lying or they’re going to cause great suffering, and a massive increase in wealth inequality.
Energy usage honestly never seems to be a concern for people, so I don’t even try to make that argument.
While I understand new data enters for ai are increasing power usage, it’s just highlighting the existing problems where there are decades of insufficient investment in infrastructure.
You can’t get enough power to run a new data center? Where were you when I complained we needed additional transmission lines to keep bringing more renewable energy online? Where were you when I wanted the huge infrastructure project to import huge amounts of Canadian hydro? I bet you wish you had that now.
Where were you when I complained we needed additional transmission lines to keep bringing more renewable energy online?
I’ve strongly argued for this in the past.
All these tech bros with AI datacenters are putting their spare couch change together to build HVDC lines across the continent, right?
Here’s a piece I wrote to explain my apprehensive stance on AI to friends and colleagues: https://blog.erlend.sh/non-consensual-technology
What do you normally say that you’re worried sounds like an “Ai vegan”?
I just mentioned to a friend of mine why I don’t use AI. My hatred towards AI strives from people making it seem sentient, the companies business model, and of course, privacy.
First off, to clear any misconception, AI is not a sentient being, it does not know how to critical think, and it’s incapable of creating thoughts outside from the data it’s trained on. Technically speaking, a LLM is a lossy compression model, which means it takes what is effectively petabytes of information and compresses it down to a sheer 40Gb. When it gets uncompressed it doesnt uncompress the entire petabytes of information it uncompresses the response that it was trained from.
There are several issues I can think of that makes the LLM do poorly at it’s job. remember LLM’s are trained exclusively on the internet, as large as the internet is, it doesn’t have everything, your codebase of a skiplist implementation is probably not going to be the same from on the internet. Assuming you have a logic error in your skiplist implementation, and you ask chatGPT “whats the issue with my codebase” it will notice the code you provided isn’t what it was trained on and will actively try to fix it digging you into a deeper rabbit hole then when you began the implementation.
On the other hand, if you ask chatGPT to derive a truth table given the following sum of minterms, it will not ever be correct unless heavily documented (IE: truth table of an adder/subtractor). This is the simplest example I could give where these LLMs cannot critical think, cannot recognize pattrrns, and only regurgitate the information it has been trained on. It will try to produce a solution but it will always fail.
This leads me to my first point why I refuse to use LLMs, it unintentionally fabricates a lot of the information and treat it as if it’s true, when I started
A discussion in good faith means treating the person you are speaking to with respect. It means not having ulterior motives. If you are having the discussion with the explicit purpose of changing their minds or, in your words, “alarming them to take action” then that is by default a bad faith discussion.
If you want to discuss with a pro-AI person in good faith, you HAVE to be open to changing your own mind. That is the whole point of a good faith discussion - but rather, you already believe you are correct, and are wanting to enter these discussions with objective ammunition to defeat somebody.
How do you actually discuss in good faith? You ask for their opinions and are open to them, then you share your own in a respectful manner. You aren’t trying to ‘win’ you are just trying to understand and in turn, help others to understand your own POV.
Chiming in here:
Most of the arguments against ai - the most common ones being plagiarism, the ecological impact - are not things people making the arguments give a flying fuck about in any other area.
Having issues with the material the model is trained on isn’t an issue with ai - it’s an issue with unethical training practices, copyright law, capitalism. These are all valid complaints, by the way, but they have nothing to do with the underlying technology. Merely with the way it’s been developed.
For the ecological side of things, sure, ai uses a lot of power. Lots of data enters. So does the internet. Do you use that? So does the stock market. Do you use that? So do cars. Do you drive?
I’ve never heard anyone say “we need less data centers” until ai came along. What, all the other data centers are totally fine but the ones being used for ai are evil? If you have an issue with the drastically increased power consumption for ai you should be able to argue a stance that is inclusive of all data centers - assuming it’s something you give a fuck about. Which you don’t.
If a model, once trained, is being used entirely locally on someone’s personal pc - do you have an issue with the ecological footprint of that? The power has been used. The model is trained.
It’s absolutely valid to have an issue with the increased power consumption used to train ai models and everything else but these are all issues with HOW and not the ontological arguments against the tech that people think they are.
It doesn’t make any of these criticisms invalid, but if you refuse to understand the nuance at work then you aren’t arguing in good faith.
If you enslave children to build a house then the issue isn’t that youre building a house, and it doesn’t mean houses are evil, the issue is that YOURE ENSLAVING CHILDREN.
Like any complicated topic there’s nuance to it and anyone that refuses to engage with that and instead relies on dogmatic thinking isn’t being intellectually honest.
I’ve never heard anyone say “we need less data centers” until ai came along. What, all the other data centers are totally fine but the ones being used for ai are evil? If you have an issue with the drastically increased power consumption for ai you should be able to argue a stance that is inclusive of all data centers - assuming it’s something you give a fuck about. Which you don’t.
AI data centers take up substantially more power than regular ones. Nobody was talking about spinning up nuclear reactors or buying out the next several years of turbine manufacturing for non-AI datacenters. Hell, Microsoft gave money to a fusion startup to build a reactor, they’ve already broken ground, but it’s far from proven that they can actually make net power with fusion. They actually think they can supply power by 2028. This is delusion driven by an impossible goal of reaching AGI with current models.
Your whole post is missing out on the difference in scale involved. GPU power consumption isn’t comparable to standard web servers at all.
To be fair if we accidentally stumble upon fusion while foolishly pursuing AGI, that’d be a great thing
I really don’t think they do take up a lot more power but more how quickly they’re being built. At least in the US, power usage has been effectively flat, with datacenters and other growing power needs balanced with increasing efficiency…… but a lot of people want a lot of new datacenters at once
And the growth of power usage for ai data centers isn’t really all that high except that we’re structured for zero power growth. This really seems like the other side of the same issue we’ve been having with renewables: no infrastructure investment. We’ve been building effectively zero transmission lines for a couple decades. That equally means we have trouble bringing renewables online and we have trouble powering a data center that pops up. There’s every chance we already have plenty of power for the new datacenters but can’t get it there
You can think that, but you’d be wrong. A ChatGPT search uses 10x the power of a regular Google search.
Google has quietly removed their net-zero pledge from their website:
This stuff isn’t being built with renewables.
For the ecological side of things, sure, ai uses a lot of power. Lots of data enters. So does the internet. Do you use that? So does the stock market. Do you use that? So do cars. Do you drive?
There are many, many differences between AI data centers and ones that don’t have to run $500k GPU clusters. They require a lot less power, a lot less space, and a lot less cooling.
Also you’re implying here that your debate opponents are being intellectually dishonest while using the same weasely arguments that people that argue in bad faith constantly employ.
The fact that a gou data center uses more power than one that does not does not matter at all.
You’re completely missing the point.
The sum total of power usage for all non ai data centers is an ecological issue whether ai data centers use more, the same, or less power.
All data centers have an ecological footprint, all use shitloads of power, and it doesn’t matter if one kind is worse than any other kind.
This is exactly what I was trying to point out in my comment.
If I take a shit in a canoe that’s a problem. Not an existential one but a problem. If I dump another ten pounds of shit in the canoe it doesn’t mean the first pound of shit goes away.
If I dump two pounds of shit in the canoe then the first pound of shit is still in the canoe. The first pound of shit doesn’t stop being an issue because now there are two more.
You can have an issue with shit in the canoe on principle, which is fine. Then it’s all problematic.
But if you’re fine with having one pound of shit in the canoe, and find with three, but not okay with eleven, then the issue isn’t shit in the canoe, it’s the amount of shit in the canoe. They’re distinct issues.
But it’s NOT intellectually honest to be okay with having one pound of shit in the canoe and not being okay with the other two. You can’t point at the two pounds of shit and say: this abominable! While ignoring the other pound of shit. Because it’s all shit.
But it’s NOT intellectually honest to be okay with having one pound of shit in the canoe and not being okay with the other two. You can’t point at the two pounds of shit and say: this abominable! While ignoring the other pound of shit. Because it’s all shit.
Sure, because that’s a terrible analogy.
Gen AI data centers don’t just require more power and space, they require so much more power and space that they are driving up energy costs in the surrounding area and are the data centers are becoming near impossible to build.
People didn’t randomly become “anti-data center”. Many of them are watching their energy bills go up. I’m watching as they talk about building new coal plants to power “gigawatt” data centers.
And it’s all so you can have more fucking chat bots.
When a family in the global south uses coal to cook their food, they release CO2. When a billionaire flies around the continent on a private jet, they also release CO2.
Do you consider the two to be equivalent in need or output?
Once you realize you can change your opinion about something after you learn about it, it’s like a super power. So many people only have the goal of proving themselves right or safeguarding their ego.
It’s okay to admit a mistake. It’s normal to be wrong about things.
Maybe part of the answer is to not be so strictly against it. AI is starting to be used in a variety of tools and not all your criticisms are valid for all of them. Being able to see where it is useful and maybe you even find it desirable helps explain that you’re not against the technology per se.
For example Zoom has an ai tool that can generate meeting summaries. It’s pretty accurate with discussions although sometimes gets confused about who said what. That ai likely used much less power, might not have been trained on copyrighted content
All current AIs are based on stolen content.
– and are being used by rich sociopaths to replace the very people that made that content.
If nothing is taken from anyone and no profit is made from a model trained on publicly accessible data - can you elaborate on how that constitutes theft?
Actually - if 100% copy righted content is used to train a model, which is released for free and never monetized - is that theft?
People downloading stuff for personal use vs making money off of it are not the same at all. We don’t tend to condone people selling bootleg DVDs, either.
Publicly accessible does not mean it is free of copyright. Yes, copyright law in it’s current form sucks and is in dire need to get reformed, preferably close to the original duration (14+14 years). But as the law currently stands, those LLM parrots are based on illegally acquired data.
Publically accessible does not mean publically reusable. You can find a lot of classic songs on YouTube and in libraries. You can’t edit them into your Hollywood movie without paying royalties.
Showing them to an AI for them to repeat the melody with 90% similarity is not a free cheat to get around that.
This is in part why the GPL and other licenses exist. Linus didn’t just put up Linux and say “Do whatever!” He explicitly said “You MAY copy and modify this work, but it must keep this license, this ownership, and you may NOT sell the transformed work”. That is a critical part of many free licenses, to ensure people don’t abuse them.
If nothing is taken from anyone and no profit is made from a model trained on publicly accessible data - can you elaborate on how that constitutes theft?
Actually - if 100% copy righted content is used to train a model, which is released for free and never monetized - is that theft?
Cool. So you’re in support of developing a model that financially compensates all of the rights holders used for its training data then?
Yes, I am. But i don’t expect them to do that.
Sort this one with the girlfriend’s “would you still love me if I was a worm” philosophy. It’s so far outside of reality it’s not worth considering.
You mean commercial LLMs.
AI as a term includes machine learning systems that go back decades.
I want my creations to be precisely what I intend to create. Generative Ai makes it easier to make something at the expense of building skills and seeing their results
Very simple.
It’s imprecise, and for your work, you’d like to be sure the work product you’re producing is top quality.
Just do what I do and say that you think it’s hot garbage that dehumanizes everything and everyone that use it.
Then go on to not give a shit what they think about it.
I’m just honest about it… “I don’t find it useful enough and do find it too harmful for the environment and society to use it”
And you then spend longer verifying the information its given you than you would have spent just looking it up to begin with.
