Mistral’s new “environmental audit” shows how much AI is hurting the planet
Individual prompts don't cost much, but billions together can have aggregate impact.
https://arstechnica.com/ai/2025/07/mistrals-new-environmental-audit-shows-how-much-ai-is-hurting-the-planet/?utm_brand=arstechnica&utm_social-type=owned&utm_source=mastodon&utm_medium=social

We are now encouraged to use Copilot AI app in work. I prefer to learn and use my brain so I pretend I use it. In my latest 1-2-1 my manager said it's great I'm 'on side' and I'm a role model for my quality of work & embracing the change in a positive way. Guess that was from AI

That final quote was from Sean Goedecke https://www.seangoedecke.com/after-the-ai-bubble/ where the full piece is overall more upbeat about the current AI boom.

I was talking to someone yesterday (let's call them A) and they had another "AI" experience, I thought might happen but hadn't heard of before.

They were interacting with an organization and upon asking a specific thing got a very specific answer. Weeks later that organization claimed it had never said what they said and when A showed the email as proof the defense was: Oh yeah, we're an international organization and it's busy right now so the person who sent the original mail probably had an LLM write it that made shit up. It literally ended with: "Let's just blame the robot ;)".

(Edit: I did read the email and it did not read like something an LLM wrote. I think we see "LLM did it" emerging as a way to cover up mistakes.)

LLMs as diffusors for responsibility in corporate environments was quite obviously gonna be a key sales pitch, but it was new to me that people would be using those lines in direct communication.

@gsuberland

New from 404 Media: a startup is selling data hacked from peoples' computers to debt collectors, divorce lawyers, more. People already hacked, now being re-vicitmized by startup. I used the tool, found peoples' personal addresses.

“This is so gross and predatory.”

https://www.404media.co/a-startup-is-selling-data-hacked-from-peoples-computers-to-debt-collectors/

Fascinating poster from the American Holocaust Museum. Thank goodness we aren't exhibiting any of these right now.

An incomplete list of people excluded from 'international' conferences that are in-person only and aren't streamed or recorded
- People with caring responsibilities
- People with disabilities
- People with health vulnerabilities
- People who don't fly in recognition of climate emergencies
- People who work for orgs with ethical no-fly policies
- People who have their travel disrupted due to climate chaos
- People who only engage with fully open access scholarship (if you have to travel+pay to watch a talk, it's not at all open access, even if a related publication is)
- People who can't leave their country for fear of not being let back in
- People who can't get visas
- Unfunded/underfunded people
- People with unfavourable exchange rates
- People who have limited holidays off work
- People who can't afford it
- etc..

I guess you end up with rooms mostly of academics with travel budgets who are too senior to really do research any more, and research students with stipends. That's fine but the perspectives can feel limited, ungrounded and underinformed.. with a huge environmental cost.

When using `remotes::install_github()` non-interactively, e.g. creating Docker images, package dependencies will all be upgraded by default (unless the R_REMOTES_UPGRADE environment variable is set).

If that variable is not set, the upgrade argument will be "ask". However, in a non-interactive environment, "ask" is equivalent to "always", which is to always upgrade.

Therefore, set upgrade to "never" if you don't want your dependencies upgraded.

https://remotes.r-lib.org/reference/install_github.html #rstats

One of the problems with vibe coding is that the hardest part of software engineering is not writing the code, rather it's *choosing* what to code, and designing the system (and, later on, maintaining the code/operations/etc)

The barriers and investment cost to writing code is itself a *desirable* aspect of software engineering because it forces you to make careful, good choices before you invest in building something

Because the majority of the time spent writing, say, curl, is not writing the original tool but rather maintaining it over time, it's important to make good choices from the beginning, and at every major version change

Public service announcement.

(Created by https://bsky.app/profile/campfireharve.st )

Can we now please stop saying (or thinking) that the public sector will never be able to produce #LLMs that can compete with big tech…especially when they are focused on the #publicgood, are built on each other’s work using public funding, and have integrated #responsibleTech principles from the very beginning…

🟣 Model with 70 billion parameters
🟣 Trained on 1500 languages
🟣 Trained on 60% English, 40% non-English
🟣 Open Source Apache 2.0 license

🇨🇭#ETHZurich #EPFL ❤️

https://ethz.ch/en/news-and-events/eth-news/news/2025/07/a-language-model-built-for-the-public-good.html

Boosting is really important on the Fediverse because it makes whatever you boost federate to the entire servers of all your followers. A boosted public post appears on your followers' timelines, but also becomes searchable to everyone on the servers of your followers.

Instead of an algorithm, the Fediverse relies on human beings sharing stuff they find interesting. This process creates a wonderful chain of discovery.

To boost something here, click the 🔁 or 🚀 button below the post.

#FediTips

Please join me in also signing this important open letter initiated by Australian academics: https://openletter.earth/an-open-letter-from-educators-who-refuse-the-call-to-adopt-genai-in-education-cb4aee75

“We will not contribute to the erosion of academic freedom and educator agency by forcing educators into compliance with technology they find unethical.” -- Miriam Reynoldson and Signatories (2025)

This was a brave project with an ambitious goal and I think it kicks the legs out from some shaky, simplistic assumptions about cycle time, and how we use software metrics in general. We've also got a huge bank of code, statistical approaches, and methodology along with our preprint so others can replicate our work or use it as a learning example.

https://arxiv.org/abs/2503.05040

It makes me laugh/cry that we spent decades trying to get the software industry to internalise that it takes far more effort to support & maintain systems than it does to write them in the first place, and yet seemingly every trendy development in the last 5-10 years has been about making that initial stage faster & sloppier at the expense of everything else

Google continues the industry-wide trend of jamming AI down users' throats, making it difficult or impossible to opt out, and potentially endangering the privacy of communications: https://www.neowin.net/guides/google-can-now-read-your-whatsapp-messages-heres-how-to-stop-it/

Someone's kindness made me cry today.

My son's been attending a kids sewing class at a local fabric store.

The owner saw how much he loves it, and had offered to sell us a machine at a discount for him to use at home, but it was still out of reach for us financially. I told her we'd save up for it for Christmas.

Today she gave us that sewing machine free. She said she wanted him to be able to continue to sew at home.

Enabling his passion is such a priceless gift and I'm incredibly thankful.

If you're like me and spend a lot of time in the terminal, check out https://github.com/johnkerl/miller!

> Miller is like awk, sed, cut, join, and sort for name-indexed data such as CSV, TSV, and tabular JSON.

Just download a static binary from the repo and include it in your PATH.