Gabriel

It seems like berries really are the LLM model’s weakness.

Human just outside the loop.

i unironically think that the richard scarry pig who is a butcher who is also selling bacon and other pork products is an excellent metaphor for a lot of what is going on right now tbh

A challenge with AI adoption is that organizations are not built to a Grand Plan where AI can just be slotted in by leaders, but rather socially constructed, semi-random & in flux

Here's an anecdote from a paper on how a CEO realized he didn't understand how things really worked (& that nobody did)

The thing that stands out to me about the latest code IDEs (windsurf, cursor, etc) isn’t that they get themselves into trouble - it’s that they can (and often do) get themselves OUT of trouble.

I have literally never seen a graph like this once in all my years of being an election sicko

18-24 always lags, always, it’s just a natural law of politics

and this is the guy they want to throw out of the party?!

🤣🤣🤣

The Misleading Panic over Misinformation People can’t agree on how to define misinformation and its related terms, making regulation nearly impossible. Instead, free expression serves as a better tool for discovering truth.

🧵 THREAD: Why the Cato Institute’s new paper on “misinformation panic” is dangerously wrong, and why it completely misunderstands the real crisis we face.
www.cato.org/policy-analy...

Crash (exploit) and burn: Securing the offensive cyber supply chain to counter China in cyberspace If the United States wishes to compete in cyberspace, it must compete against China to secure its offensive cyber supply chain.

I’m logging on from vacation in Tokyo to share this banger of a report comparing the cyber offensive acquisition strategies of China and the US.

www.atlanticcouncil.org/in-depth-res...

Introducing AIRTBench, an AI red teaming benchmark for evaluating language models’ ability to autonomously discover and exploit AI/ML security vulnerabilities.

Read the paper on arXiv: arxiv.org/abs/2506.14682

Open-source dataset and benchmark eval code repo: github.com/dreadnode/AI...

Infosec - but especially threat intel is a team sport

A singular act of defiance that will never go out of style.

This week's show is live on all platforms (Mikko Hypponen fills in for Juanito) @craiu.bsky.social @jags.bsky.social @mikko.bsky.social
securityconversations.com/episode/mikk...

Have LLMs Finally Mastered Geolocation? - bellingcat We tasked LLMs from OpenAI, Google, Anthropic, Mistral and xAI to geolocate our unpublished holiday snaps. Here's how they did.

Bellingcat has followed up its 2023 article on testing LLMs on geolocation with a new review of LLM's geolocation skills, with dramatically improved results www.bellingcat.com/resources/ho...

If it cures the anxiety… 🤔

One positive outcome of "vibe coding" becoming widespread across devs is that prototyping becomes widespread.

Until now, it was accepted that for most engineers, prototyping is kind of hard? So I saw relatively few do it.

Now: SO easy with these AI tools! Fewer excuses left?

Incredible tool to collect data for a finetune.

👀…

Whoever you are.

I see you LYING in your metadata and messing up my data.

I'm not mad, but I want you to know that I am writing a parser just. for. YOU.

Bigger isn't always better: how to choose the most efficient model for context-specific tasks 🌱🧑🏼‍💻 A Blog post by Sasha Luccioni on Hugging Face

How can we make informed choices based on performance AND energy when using AI in real-life tasks like question answering? By evaluating them and picking the models that optimize both factors!
Check out my new blog post on the subject:
huggingface.co/blog/sasha/e...

God cursed me with a lot of stuff but he did bless me with thinking oatmeal tastes absolutely decadent

Tweet by Sam Bowman @sleepinyourhat If it thinks you're doing something egregiously immoral, for example, like faking data in a pharmaceutical trial, it will use command-line tools to contact the press, contact regulators, try to lock you out of the relevant systems, or all of the above.

welcome to the future, now your error-prone software can call the cops

(this is an Anthropic employee talking about Claude Opus 4)

"You read books, eh?" A 1949 red scare Herblock cartoon about pressures on teachers that hits hard once again.

wife: how was guarding the two paths today, honey?

guard: [looking away] fine

wife: did something happen?

guard: [tearing up] no

wife: would the other guard tell me something happened?

Imagine starting a car that hadn't run in 21 years, that's 15 billion miles away in interstellar space. That's what the NASA team just did with Voyager's thrusters. People are amazing. jpl.nasa.gov/news/nasas-v...

Will my skills pay the bills?!?

And by bills, I mean mortgage.

And by skills, I mean vibe coding.

Masked, armed ICE agents arrest two men in Great Barrington apartment building as witnesses taunt, shoot video The arrests this morning in Great Barrington were witnessed and videotaped by multiple bystanders in The Triplex Cinema parking lot. Tenants of the building say ICE has surveilled the building

A key element of the administration’s repurposing of wartime and counterterrorism framings to other ends is aesthetic.

Here ICE agents are kitted out like wannabe GWOT operators while conducting a raid in the badlands of the Berkshires.

www.berkshireeagle.com/breaking/ice...

There's a 2021 stealer log for a telemessage employee's hootsuite login. Other websites in the same stealer log (looks like websites the same machine visited) include readcomiconline[.]to and kissanime[.]ru. For national security reasons, Anime was a mistake

Credential revocation API to revoke exposed PATs is now generally available · GitHub Changelog You can now revoke an exposed GitHub personal access token (PAT) you found outside of repositories, even if it’s not yours, to help quickly limit the impact of the exposure…

This is incredible news! Every API should offer this type of endpoint. If someone finds a leaked API key, let them report it back through the API safely.

"Credential revocation API to revoke exposed PATs is now generally available"
github.blog/changelog/20...