In particular I’m not sure if my “woke cat” comics should have the box checked for future posts.
Written by me, and then I take multiple chat bot images and make manual corrections and edits in GIMP to get the comic panels.
So if I had to pick yes or no to “is it AI generated?” I’d say “no”
But I’m not sure if the “AI generated” check box is equivalent to that question, or more like “does it contain auto generated content?” which would be “yes”
Edit - will absolutely not be using the check box for these comics due to a fucked up reply that got too many upvotes
It is not a meme, it is happening and i know that you work there and are probably the most knowledgeable person among us.
However, you can’t dimiss that they do use ressource as water, rare earth…The problem is that they benefit outrageous funds and double their calculation power each year.
That’s not about the result but the road to learn through hardship how to create something. It create a feeling of fullness. If you give yourself more time, you could achieve a good work instead relying on a third party and tell your kid “i did it”.
You really think they never used it to change our history narrative ? It is not something new, the Trajan column in Roma serve the same purpose.
The french “resistance”. They removed the video since but it was published by our government :
China trying to take over Taiwan :
Did not realize it was you :-) I don’t want to be aggressive, just to dispel some myths that I really think are harmful to understand this field (in which I am working) well.
My image generation and local LLMs do not use water. They are air-cooled.
Datacenters do use water (during operation) and rare earth (during construction). I am disagreeing on their use being unsustainable.
You know, it is not because I work in the field that I defend it. If these claims were true, I would simply have not chosen to work in the field. I remember very well in engineering school when I first was confronted over the claim that some mineral resources usage were unsustainable. It was copper. It shocked me. Copper, so central to electronics, which I was studying, was going to deplete. Damn. “How much left do we realistically have?” was the first immediate question. Is that solely an engineers instinct to try to quantify problems? Once I dug up, I realized that the claims were untrue. That reserves were extremely abundant and that the people raising the alarms simply did not understand geological reports. It was in the late 90s. Since then we have “ran out” of copper several times according to doomsayers.
“Sustainability” can mean several things and is used in different definitions when considering different resources. Oil usage for energy production is the text book example of unsustainable process: we need to destroy at a given rate a resource with a limited stock to maintain an activity.
Water usage in datacenter is not unsustainable according to that definition: water is not destroyed, the stock is not depleted. Datacenter implementations can be problematic in places where not enough clean water is available ofr the population but it is not an ecological problem or a resource exhaustion problem. It is an infrastructure problem, and it comes with a ton of political and social consideration. But in that respect there is not a lot of difference between implanting a datacenter somewhere and putting a farm, that will also use water.
Rare earths are used for construction. Building a datacenter requires a set amount of it, but operating it does not. One could argue that infinite growth of the number of datacenters is unsustainable, which is true for almost any infinite growth, but no one is believing that we need to double the number of GPUs every year for centuries.
Also, rare earth are badly named. They are not rare. There are plenty of deposits, plenty of reserves and none of these are predicted to be in geological shortage. The only one you could make a case for is helium, which is not used in AI datacenters. The USGS is a good resource for quantifying the scarcity fears.
Here again, conflating many issues. Image generation models were used since day one for problematic uses: deepfakes, revenge porn, propaganda. Same can be said of photoshop (and no, making a goof convincing fake with AI is not easier than with photoshop. I remember the exact same moral panics when I was a teenager and some people were copy-pasting stars or classmate faces over nudes with photoshops).
What I am arguing is that we have no example so far of image generation models purposefully trained to follow an agenda. They have biases (they will gender and racialise many profession for instance) but these biases come from their datasets, “what they see”, not from an agenda. On the contrary, the examples you give, Grok (from v3) and Deepseek, were demonstrably trained with an agenda in mind. (If you have sources about OpenAI being caught doing the same, I am interested, I may have missed it)
LLMs don’t have to be trained that way. It is not inherent to the tech. Many models are uncensored. Open weight models like DeepSeek have been also modified by the community to remove their censorship (with occasionally hilarious effects, as a model trained to “never do X” is often easy to reversed to “being very good at doing X”)
Actually the layer of fine-tuning that these agenda-givers use, is a technically simple but very political process that I wish more people joined instead of lambasting AI as a whole.
If you have good ressource on ecological side of AI and rare earth, i’m interested because i don’t have lot informations. Most of the tell us that they are building nuclear reactor. You may want create a post and invite use to raise our knowledge on these topics.
Most of my knowledge come from the ADEME (french ecological structure that analyse various situation), and they did predict metal ressource depletion around 2050.
More exactly, digging machine will have a lower mining rate for a given ressource. We will have to remove lot more earth to find iron, copper… it’s the same problem for oil.
And for the misuse of AI, that’s the scale of it that worry my the most : governement and big tech. Especially big tech, i’m fearfull of them. I would prefer a collective solution : made by people, for people and respecting human diversity and right.
I haven’t given example of OpenAI but Microsoft is one of its main investor. They were at Trump meeting with other american big tech. So my trust that was already low dropped to the bottom. 😅
The USGS is my go-to source for mineral reserves. Their website is becoming harder to navigate by the year, here is a directory with their reports on several minerals: https://pubs.usgs.gov/periodicals/mcs2025/ And also here is a mandatory reading on the various definition of ‘reserves’ that they use. The misconceptions usually come from there.
I am too. Because I always see the claim, “there is a rare earth problem with AI” but every time I ask for specifics, I get nothing. I have had to defend against the claims of depletion of copper, lithium, uranium, gallium, in the past, I am ready to defend whatever they now claim is going to miss but it always remains vague.
Rare earth are strategical and of interest to states but not because of mineral availability reasons, because of capitalist incentive reasons. What people do not realize is that the actual problem is that these minerals are actually very cheap, yet hard to extract! They are usually the byproducts of other mining (IIRC gallium is a side product of some aluminium deposits) and are absolutely not profitable to mine solely by themselves. Yet, if you don’t have access to a source, no microchips (for gallium) or efficient electric engine/turbines (for neodymium) for you!
So there is an incentive to have access to mines, but the precious part is the actual mine with its investment done and its working business, not the geological resource.
I am a bit tired of defending it online but maybe I should indeed. I have tried to make some posts on [email protected] in the past, but the hostility is draining. I feel I should spend more time helping the OSS scene rather than addressing vocal critics online that may actually be a small minority.
ADEME is usually a good source. I am surprised they would make such a claim. Do you have a link?
Not really. Oil and in more general fossil resources are special in that they come from an organic process that makes them available only on a thin layer of the crust and only where some precise geological conditions were met. Their scarcity is much more immediate than minerals which are basically available pretty homogeneously once you start digging.
Minerals do get a bit harder to get but oil becomes impossible to get at one point. Like always, the key is to quantify the problem. A (imperfect) proxy to the difficulty to obtain minerals is market price. Minerals like coppers have seen their price rise quite a bit but rare earths like gallium have seen their price go down. None seem to have an unsustainable trend.
Me too! That’s why I try to fight the rejection of “AI” as a whole. This work is currently being done, open source is actually winning this important battle in the general indifference, and it really breaks my heart to see it thrown in the same bag as some genuinely evil people of the corporate AI world.
Oh yes totally, I don’t trust OpenAI one bit, it is just that I do have examples of propaganda training by DeepSeek and Grok and would have loved to know if some proofs exist for OpenAI (or Anthropic). I don’t consider them ethical either even if they don’t, they have many other problems.