Now that my local agents can do cool stuff, having only three simultaneous API calls to really work with at once is weirdly limiting Jevons paradox am I right? Maybe I should consider bumping down to Q4? 😬 Nah who would do that

Sweden and Japan looking pretty good

If you, human, trusted an AI hallucination isn't that really your fault?

Why even bother having a legal framework, governance systems, etc, if in the end it's just 'well we all know that guy was bad, and you're in love with the law so by proxy you love him but we had to do it, we just had to. He's a bad guy you know? B A D M A N, you see? Had to do it, no other options!'

Fetterman.gguf needs some temperature tuning or something, dry needs to be a little more aggressive too. Philosopher kings 🤡

this is a good way to phrase it not to dismiss it, it’s just that it’s happening whether it’s moral or not. might as well focus your energy on finding ways to help people make it through with their soul and dignity in tact

Llama.cpp update, 3 more tps 😎

So I haven't really been using openwebui much recently, distraction and all - but it has implemented a bunch of new features that seem pretty good and in line with trends. In particular, open terminal is an exciting development I'm testing with Qwen3.5 now

I wrote about how Democratic leadership has abandoned the moral high ground on Iran, and how it has left them toothless when opposing Trump.

"these were the quaint sorts of posts Doll made in the early days, before they were the powerful agentic tycoon heading Dollcorp"

Yes. They made a bad mistake, and they should have to pay dearly for it.

I slept about an hour last night, so I'll be more unhinged than usual

In high school I think I always did badly in things like calculus because I literally never could connect it with what I'd do with it. Teacher didn't have time, or ability to explain why it mattered. Later on when I was in economics I could appreciate it, when it was *applied*

This, but As soon as [profession] feels the power of the AIs encroaching, they will start coming up with lots of reasons why they are smart and good and the AI isn't.

Mildly annoyed that I can't do tensor parallelism on 3 gpus; but we're up at around 37 ish tps on average, on Q5 Qwen35... Which is fine. Usable. It's usable 😬🤬 (Full 256k context allocated, up to 3 concurrent requests for sub agents)

It was all about spying on Americans: www.theatlantic.com/technology/2...

--dangerously skip international law

look, there’s no reason not to cancel your OpenAI subscription now

Look, if you believe the government or *the fucking Pentagon* when they tell you they aren't surveiling you, mass-ly You're nothing but a sucker. 4th amendment basically doesn't exist anymore, thank the Supreme Court for that - though many politicians, including left leaning are also culpable.

Thinking about trying something like Gauntlet of 20 domain specialized questions -> llm Copy thinking + solution > 20 files Frontier LLMs harshly examine reasoning and conclusions, and determine the flaws in the thinking here specifically - 3 pass different models > 3 assessments

The scene at the end of every Rick and Morty season where Rick is going on about "every day Rick and Morty, 50 years of Rick and Morty" But it's me with solar

AI replacing all work isn’t a “doomer” scenario, it’s a utopian vision of a post-scarcity world in which it would finally be possible to enact every socialist dream where everyone has high living standards and freedom from want or oppressive toil.

Interesting the areas in which the denser 27B model is basically equal to the 35B A3B model - but in any case, it's now easy to run a OSS-120B at home; and GLM 4.5-air is just blown out of the water. I remember being really jealous of people who could run 4.5 air well - 5 ish months ago.

Already back up to 45 tabs 😬

I can cosign this

Not to excuse anything but *maybe*... The logic is - "if we all banded together and then lost at the supreme court, admin would nationalize the labs and then the world would be truly fucked" If one lab is cooperative, temptation to nationalize is lower in theory But that's probably not it

Yeah, this is unacceptable. Generally I do like open AI's models - but now they've given me a reason to turn away from them. Stand with anthropic on this, or you're scum.

Cursed, but I laughed.. Also, I don't know what this wetness is people keep talking about

I am actually surprised that anthropic isn't caving, that's what tends to happen; very commendable, maybe we can do this after all guys.

lol that you can't force someone to make a gay cake in this country, but they can force you to make a weapons targeting system

Qwen35 is having no problems spawning sub agents - the only problem now is I gotta figure out some good ways to use them. So far, very exciting results.

anonymity is probably dead, focus on privacy simonlermen.substack.com/p/large-scal...

You’re Absolutely Right! Let me neutralize all hostile targets in this combat zone before moving forward ☺️

Liquid has made some really good small models - between this and Qwen3.5 this is going to be an interesting weekend

It's depressing that there are no principles anymore

someone offered to buy me crystal Shrek. i said no because I am a moral person. but it did take me a full beat. there is sin in my heart.

I've been going through the backlog and I know I'm a broken record here But you should just drop what you're doing and listen to the 5-4 episode on Qualified Immunity

When I tried online classes in college I hated them and did worse, fwiw. That was like a century ago though

The US Justice Department withheld testimony that Donald Trump sexually abused a girl when she was 13 or 14 - according to a source who viewed unredacted Epstein files. This is the worst scandal in US history and should be the biggest news story on the planet right now. www.npr.org/2026/02/24/n...

AI is the greatest threat to liberty today and I dont even think its particularly close. I'm excited about AI but deeply concerned about its centralization and governance. Glad anthropic is (for now) doing the right thing. When they go public they will likely lose the ability to do this.

Wyd

Generally, if you are technical enough to get openclaw set up you should probably be technical enough to understand and mitigate the risks; you may choose not to do that, but it's at your peril right Also, the idea that this recent event is somehow an own on AI people is very dumb

wow nature is beautiful

The "STOP OPENCLAW" story is just actually hilarious. The desperation! Incidentally, I don't use openclaw but it seems totally doable to just add a /Halt command that immediately stops all execution on a programmatic level.