Helpful reminder that data center problems are not just hypothetical—they’re potentially local.

Closing digital divides is good, and increasing diversity in tech fields as good, but I’ve been complaining for years about computer science ed that we stop at the nobility of those goals and don’t ask ourselves about the deeper motivations behind those initiatives. So it is with AI: A more diverse field more available to all is better than what we have, but we also have to ask whether AI education is actually a social good.

This morning on Bluesky, I saw some posts about a class action lawsuit against Anthropic for their use of pirated, copyrighted materials in training their generative AI models. One of the sources of these copyrighted materials was the LibGen database, which I took a peek at nearly six months ago to confirm what I was already sure to be true: that my scientific writing was also collected as training material by companies like Anthropic or Meta. I don’t love that big tech companies are profiting off of my work in this way, and I’m sympathetic to the authors who are taking legal action against Anthropic. However, as I’ve written repeatedly over the past few years (you can find some of those thoughts—and others—by scrolling through here, I don’t know that copyright is the right way of responding to this kind of abuse.

I genuinely think it’s useful to remember that non-AI datacenters are also contributing to the climate crisis, but that doesn’t let AI off the hook. It’s like saying “sure, we’re spending far beyond our means, but have you considered that we’re already in debt?

This sucks. I don’t have a lot of sympathy for Reddit here, which has shown over the past few years a dedicated interest in monetizing its userbase.

Ma carrière se divise entre une valorisation des pratiques numériques perçues comme n’ayant pas d’importance et une critique des technologies qui permettent ces pratiques. J’avoue que je me sens vraiment divisé pour cet exemple !

Look, nothing really new in here (Clearview should have made parents rethink the same ages ago, etc.), but yes, AI should get parents to be a hell of a lot more careful with posting pictures of kids.

I hadn’t thought about these two trends (cracking down on adult content, and Grok being Grok) being in tension with each other, and I appreciate what this article does to make that clear.

Unsurprising but disappointing.

Again, I’m not sure copyright is the way to go in fighting immoral generative AI companies (that the ALA and EFF are on Anthropic’s side seems important to me), but “we have to be able to do this to be successful” still strikes me as such a hollow, self-serving argument.

À mon avis, l’expertise « au niveau doctorat » est surtout lié au processus de connaître et non à la connaissance elle-même, et l’IA générative ne respecte pas du tout ce processus.

This seems performative to me, and this paragraph gets at why I think so:

AI companies are increasingly pushing into education — perhaps in part to try and fight the reputation that AI tools have acquired that they help students cheat. Features like Gemini’s guided learning mode and ChatGPT’s similar study mode, which was announced last week, could theoretically help with actual learning, but the question is whether students will want to use these modes instead of just using the AI chatbots for easy answers.

There are, of course, social benefits to open sourcing powerful tools like these ones. However, I’m reminded of “open source” Android, which is a deliberate business decision that benefits Google—and of how many NCII-generating tools are based on open weight/open source models. gift link

I’m glad that someone is doing this white hat work, but I hate that we live in a world where someone has to.

Quel monde pourri qui attend nos enfants.

Kind of hate that we have to ask the question in this headline!

I’ve been (link)blogging recently about needing to talk about epistemology when we talk about generative AI. I know that in at least one case, I’ve evoked the generation of scientific knowledge as a counterexample to the “just the facts, ma’am” naïve epistemology promoted by AI and its supporters. To use generative AI to review papers—and to try to get around peer review—feels particularly dangerous to me.

Quick question about this passage:

Trump … signed an executive order titled “Preventing Woke AI in the Federal Government,” directing government agencies “not to procure models that sacrifice truthfulness and accuracy to ideological agendas.”

So, how does one determine what is true or accurate? Once again, we need to talk about epistemology when we talk about generative AI.

Dammit, am I going to have to stop using Canvas?

I don’t think copyright is the best argument against generative AI (strengthening copyright law will benefit big companies more than small creators), but “can’t make an AI omelette without breaking a few copyrighted eggs 🤷🤷🤷” is still a depressingly cynical national policy.

This emphasis on “objective truth” further underscores the need to talk epistemology when we talk AI.

Je n’aime pas du tout la présence des IA dans les écoles, mais je trouve aussi gênante la pénalisation à tort des étudiants.

You know, I skipped over this story when it come out in a couple of other outlets, but seeing the headline again here got me thinking about how good/scary of an example this is of LLMs shaping (rather than reflecting) reality.

This sounds fine.

As a committed em dash user, this has been bugging me since I heard about it.

Je compte bientôt préparer un document professionnel en français. Il y a cinq ans, j’aurais demandé à quelqu’un de m’aider à corriger des erreurs. En 2025, par contre, je me demande si mes petites erreurs éventuelles feront preuve que j’ai tout écrit moi-même au lieu de me fier à l’IA.