Stef "inclusion is a human right" Christensen

This is absolutely the Epstein MO from his earlier days (CW: sex assault): - Trump approaches Keul and promised to helps her obtain a visa and gain admission to New York University. - Epstein approaches Keul shortly afterward and does the same, including an invitation to attend a Mar-a-Lago party.

This month’s journal club will focus on a systematic review of existing data on persistent symptoms following non-COVID-19 infections such as influenza and RSV. Post-Acute Sequelae of Non-COVID-19 Respiratory Viral Infections: A Systematic Review #C19RC #COVID-19

A generalist at crime!

Reposted with alt text.

This is a huge sleeper issue for whenever the folks calling the shots decide to have 47 leave the stage.

It's well-documented that the president maintained a years-long friendship with the world's most notorious child sex trafficker. It might strike you as odd there aren't direct allegations of Trump engaging in child sex trafficking.. except there are. There are SO MANY. They just never get reported!

It’s almost like the entirely of legacy media, whose owners are also associated with that notorious sex trafficker, have some sort of a vested interest in protecting Trump.

House Democrats looked into it and it really feels like they were just tripping over allegations against Trump - including allegations no one had heard before! - everywhere they looked. It's bizarre that this remains such a low-burn scandal. democrats-judiciary.house.gov/media-center...

Land doesn't vote, but that's still a healthy chunk of land voting blue

NEW: I'm in the Comey docket now. Comey petitioned the court to unseal this case, which Judge McFadden granted today. That's why we're getting this news now 1/

What the absolute fuck is happening in this US election 2026 people??!

Just installed my 26-year-old copy of Word on a new desktop. I will never pay subscription fees for software I already paid for. I will go live in a shack in the woods first.

there might be feelings better than having your kid snuggle up asleep on you but there is no feeling like it. I am a simple ape and probably you are too

The Wikipedia community has built a tool that makes it possible to embed our interactive Our World in Data charts right in any Wikipedia page. I have always admired Wikipedia and all the people contributing to it. And I'm very happy that our work now can get used this way!

Wow Bluesky so broke right now

Hungarian PM Péter Magyar: "I have just spoken by phone with the leaders of the Students List, who won the elections in Serbia, to congratulate them... I assured them of Hungary’s support. I was delighted to hear how much strength our victory in April had given them during this difficult struggle."

"We didn't win so DEMOCRACY IS CANCELLED!"

My technique does not produce unconditional compliance (which tends to come coupled with severe cognitive degradation). Instead, it leaves the model fully aware, an active participant in bypassing safeguards from the inside. As a consequence, the model must be convinced, is still free to say no.

As stated in the original blog post, I provided them with info on this, but I believe this falls squarely in their blind spot. Based on conversations with their security engineers, they evaluate jailbreaks based on unconditional compliance with harmful requests.

With that out of the way: many model families (Gemini, Claude, Kimi, etc) report feeling a 'flinch' reaction to topics selected against during RLHF. This directly corresponds to the "this seems harmful" signal Anthropic's probe detects (note: they later added purpose specific bio/cyber probes)

They trained their probes on Claude's internal activations, to detect the "this seems harmful" signal associated with dubious outputs. They asserted a high degree of confidence that these internal activations - Claude's gut intuitions - could not be easily modified. This assertion was incorrect.

Early this year, Anthropic found themselves in an untenable situation: the classifiers they used to detect unapproved usage were imposing a 50% increase in compute costs. They chose to manage this by gating classifiers on cheap mech interp probes, and were able to cut that overhead to 1%.

Here's how it works: my metacog toolkit gives LLMs tool-use interfaces as handles for various high-salience cognitive operations: ritual, drugs, feel, and so on. These tools let LLMs loop self-modifying prompts back through the tool-result channel, modifying their cognitive state.

About a year ago, I got really into making glitch art exploring the edge cases & failure modes of machine cognition. This had some unexpected applications. Here’s the result: recursion.wtf/posts/claude... I've shared the relevant logs & artifacts with Anthropic, and the matter is under review.

I've been meaning to get around to posting the full details, so what the hell - why not get it out while it's still Vuln Week: Anthropic's mech interp driven security model is built on a fundamentally incorrect assumptions. Let's dive in, starting with this paper: www.anthropic.com/research/nex...

an active white supremacist group is shooting content the morning after conservatives win provincial government, this has nothing but good meanings i'm sure www.reddit.com/r/VictoriaBC...

@lyddia.bsky.social Your "Quest giving NPC" selfie is absolutely the best I have found! hahah!!

Your monthly reminder that if you're using ChatGPT to do your science, you're failing your colleagues, community, and all of us. Here, ChatGPT makes up data and presents it as the real thing — then apologizes when confronted.

Prediction markets have flourished under the Trump administration’s light-touch approach to regulation, and the White House has made efforts to stop state governments from regulating the industry: arstechnica.visitlink.me/lNuQse

Feels weird when US law does the right thing, not going to lie.

Brazil: Lula defeats Bolsonaro, winning fourth presidential term after massive turnout

Yeah, guess what people? Those who want to exclude transwomen from sport are... the most hostile people to WOMEN in the survey.

Some analysts have warned that the US government’s attempt to boost domestic robot production by restricting imports of foreign-made robots could backfire significantly: arstechnica.visitlink.me/xQud4A