sebastian

The so-called Department of Government Efficiency: We saved $1M per year by converting 14,000 magnetic tapes (70 year old technology for information storage) to permanent modern digital records

YOU DID WHAT?

The 2025 AI Engineering Reading List We picked 50 paper/models/blogs across 10 fields in AI Eng: LLMs, Benchmarks, Prompting, RAG, Agents, CodeGen, Vision, Voice, Diffusion, Finetuning. If you're starting from scratch, start here.

For papers have a look at this list: www.latent.space/p/2025-papers

Just remembered 'Pwned or Bot' by @troyhunt.com: using breach history as identity validation. Past pwns = digital footprints! Clever way to spot real humans in our bot-filled internet.

Developer Creates Infinite Maze That Traps AI Training Bots "Nepenthes generates random links that always point back to itself - the crawler downloads those new links. Nepenthes happily just returns more and more lists of links pointing back to itself."

“Let's say you've got horsepower and bandwidth to burn, and just want to see these AI models burn. ... It's also sort of an art work, just me unleashing shear unadulterated rage at how things are going.”

love to see it

www.404media.co/developer-cr...

As other have said Brother is a great choice.

retro.social/@ifixcoinops...
This explains why: "Brother's have remained consistently Fine I Guess, which now makes them the best printer manufacturer simply by virtue of them opting out of the Who Can Get Crappiest Fastest race"

Oh wow. That's super cool.

Five things privacy experts know about AI - Ted is writing things … and that AI salespeople don't want you to know!

This is an excellent primer on some of the privacy dangers posed by large scale AI, from a cybersecurity perspective. Written in clear language, it's the most accessible rundown I've seen yet on these topics!

desfontain.es/blog/privacy...

who is this for? that's what I can't wrap my head around - who wants to follow someone who's not real, and is posting about their regular day to day life except none of it is really happening? who is this *for*?

Volkswagen left an unprotected database with up to two years of sensitive personal data on 800k networked VW, Seat, Audi and Skoda cars accessible online, including names, user IDs, sensor and geolocation data.

CCC talk by Flüpke and @michaelkreil.bsky.social:
streaming.media.ccc.de/38c3/relive/...

💀

GitHub - OpenMined/30DaysOfFLCode: Official Repo for the 30DaysOfFLCode Challenge Initiative Official Repo for the 30DaysOfFLCode Challenge Initiative - OpenMined/30DaysOfFLCode

@openmined.bsky.social's #30DaysOfFLCode challenge just wrapped up, and they've curated this fantastic resource list for learning about federated learning. Definitely worth bookmarking if you plan to explore FL in the future: github.com/OpenMined/30...

#FederatedLearning #PrivacyPreservingAI

it depends

In case you don't have time to read the EDPB opinion on AI training, here's a summary of pretty much every paragraph.

Used to hate finding typos after hitting send. Now I'm just like "Well, at least they won't think I'm a LLM."

Quantum Country A free introduction to quantum computing and quantum mechanics

quantum.country by @andymatuschak.org.

I haven't read it myself (yet) but the way this 'mnemonic book' is laid out looks awesome!

From a legal perspective this isn't really covered AFAIK (or there doesn't seem to be an issue with it). Even if it's your identical twin brother that uses services like these.

Remember That DNA You Gave 23andMe? The company is in trouble, and anyone who has spit into one of its test tubes should be concerned.

I think @carissaveliz.bsky.social touched on this topic in her book "Privacy Is Power". Highly recommended btw.

There is also this article by The Athlantic, but it is unfortunately paywalled.
www.theatlantic.com/health/archi...

Learning the ropes: why Germany is building risk into its playgrounds Lofty climbing towers are part of trend away from total safety and towards teaching children to navigate difficult situations

German playground designers embrace risk as learning tool "even if the consequence is the odd broken bone".

www.theguardian.com/world/2021/o...

Thanks for the article.

About encrypted DNS: I'm curious about choosing servers. While encrypted DNS solves the ISP plaintext issue, what makes certain servers more trustworthy beyond different jurisdictions? (personally, I'd avoid Google's DNS given their track record with user privacy)

Is the image just black for everyone else? Wondering if there was an error uploading or if it's just my client.

"How it works" from proofofhumanity.id

I wonder if something like this could work. But it might be a too strong barrier for new users that just want to try Bluesky out without having verified users able to vouch for them...

proofofhumanity.id
(ignoring the crypto aspect, just referring at the core idea itself)

Job ad: Palantir Technologies Privacy and Civil Liberties Software Engineer New York, NY Engineering / Full-time / Hybrid

::chefs kiss::

In the UK case GDPR was enacted before Brexit and is thus valid UK law.

Thanks to GDPR they need to offer this option to EU citizen and non-EU citizen living in the EU. It also applies to some non-EU countries like Norway. I think Switzerland and UK are special cases but seem to be covered here.

So yes, everyone in the EU & UK should be able to do this.

Screen shot of a dogshit New York Times article claiming that the next big real estate trend was purchasing property in the Metaverse

Do you guys remember this fucking bullshit lmao

However, that doesn't really address the password sharing. But that might just be the "move fast and break things" of access controls.

Hanlon's razor: "Never attribute to malice that which is adequately explained by stupidity."

That being said, I nonetheless think the best approach is publishing IDs that point to the entries, not complete datasets. We cannot expect people wanting to delete their content to contact everyone who made these datasets public. The burden should be on those using the data, not the data subjects.

Other obvious downsides, beyond being time-intensive: It requires knowledge of the API or software (not obvious for e.g. social scientists) and data entropy as tweets get deleted over time, making comparisons across studies difficult.

Depending on the dataset, it is more than just a mild convenience. Extreme example is the 2016 US election dataset had 280 million tweets. With Twitter’s API limitations it would take 32 days to retrieve the full dataset. (IF all tweets would still be available)

A counter-argument is that keeping the control of data with the platform concentrates the power with the company behind it, as seen with X. Open data access enables e.g. third-party tools and federation.

So it seems that while we're losing control over data, we keep some control over the plattform.