Raymond Neilson

Does it scratch the old-school x-com itch?

YouTube video by Hooded Horse Xenonauts 2 - 1.0 Release Date Trailer | Turn-Based Tactical Alien Invasion

Wait why did nobody tell me Xenonauts 2 is about to come out (as in 1.0 all proper-like)?

youtu.be/6HI-XKVW_xE

"Freud Is All You Need" (2026)

Timeline lines up

worse news: the UCP will be utterly insufferable about it

Hey that means we can add the US to the list of state sponsors of terrorism

If you need to get used to them, the new Qwen3.5-30B-A3B isn't the worst place to start

It's been a week of war in Iran. Where are we?

-The US & Israel have accumulated tactical successes. The Islamic Repubic is not on the verge of collapse, but it is much weaker.

-Again: how to translate this into sustainable strategic success is at best uncertain.

A) Enforcing it is a mug's game, all that's doing is making sites like Discord roll out shitty intrusive age verification and making politicians talk about banning VPNs too
B) I'd like to ban social media for politicians and their staffers first

Can we offer you one (1) Canadian citizenship

Evangeline (train) - Wikipedia

Did you know that (at least part of) Canada responded to a gas crisis by large-scale mode shift to mass transit? It's true!

After Via Rail took over operation of the "Dayliner", it became a very popular way to get to Halifax from the Annapolis Valley!

en.wikipedia.org/wiki/Evangel...

If Halifax's recent budget drama is anything to go by, there's a window of opportunity for the feds to twist cities' arms to massively open up zoning, in exchange for debt swaps

We have to get moving, like, yesterday, because the world is going to shit real damn quick

bsky.app/profile/shar...

"In the Emperor's name, ENOUGH"

Still hets a neck-tingle from me every time

i really do think we need to step up homebuilding + infra work in Canada because i think the rest of the world is going to hell and Canada is going to be one of the safer harbours if we can make it one

My hot take, loosely held (bc it's been a while): people should be studying Amiens instead

Post-mortem: Unauthorized Cline CLI npm publish on February 17, 2026 On February 17th, a compromised npm token was used to publish an unauthorized version of the Cline CLI. No malicious code was delivered, and the Cline VS Code and JetBrains extensions were never affec...

don't need magic strings for that one

cline.ghost.io/post-mortem-...

> This created a prompt injection vulnerability. An attacker could craft an issue title containing instructions that trick Claude into running arbitrary code on the GitHub Actions runner.

We're gonna look back on SQL injection with so much fondness

Post-mortem: Unauthorized Cline CLI npm publish on February 17, 2026 On February 17th, a compromised npm token was used to publish an unauthorized version of the Cline CLI. No malicious code was delivered, and the Cline VS Code and JetBrains extensions were never affec...

Welcome to the new age (this one's fresh, it's not the dataset poisoning attack but it *is* yet another prompt injection)

cline.ghost.io/post-mortem-...

I'm less worried about magic strings crashing and more worried about e.g. reinforcing installing slopsquatted packages. I think it's going to get easier not harder, because exploiting crashes isn't the attack vector, and we're going to have more people using e.g. claude to set up envs...

A small number of samples can poison LLMs of any size Anthropic research on data-poisoning attacks in large language models

Rummaged: www.anthropic.com/research/sma...

Mar 4 2026 BlueSky post claiming that it explained why we bombed a school in Iran "INFORMATION THAT IS A DECADE OLD" and had A Claude chat answer as its source

Hey guys this is not a viable source. Besides the fact that any AI system being used by the government presumably has access to more updated date (I sure hope). Claude is not a person, it can't answer questions about why it did something, it confidently lies, it doesn't have access to other's state.

There's a paper (I'll have to rummage for it) from Anthropic themselves about how relatively few documents you need to compromise to poison a training set and inject a backdoor instruction to an LLM. It's in the low hundreds. And all you need is to include install commands for project setup.

Fucking A dude

Oh good I'm not the only one who does that for peanut butter jars

...where the produced artifacts went from "laughably bad" to "acceptable in some circumstances by some people". Software's only now having the arguments about provenance, epistemology, and quality that academia's been neck-deep in for two years now

One interesting (derogatory) dynamic is that so much of the recent enthusiasm is from coding and coding-adjacent spaces, convinced that recent progress there is the tip of the spear, soon to spread everywhere...but I'd argue coding is one of the *last* areas to hit that phase transition (cont'd)

Butlerian Jihad already started tbh

That's why I don't spend a lot of time (here, at least) arguing, it's lose-lose(-lose)

The "works" part of "it works" is the part which holds up both sides' arguments - I know my definition (as a skeptic) is fairly distant from the definition used by enthusiasts