Axe Ghost. On Steam!

Axe Ghost. On Steam!

@axeghostgame.bsky.social

Latent-space enjoyer. Cleverly push monsters around a grid and blow them up. Demo👉 https://store.steampowered.com/app/2994110/

1.5K followers2K following5.8K posts

A community translation can run attacker-chosen code in an exported Godot game axeghost.offprint.app/a/3mysixmpap...

axeghost.offprint.appA community translation can run attacker-chosen code in an exported Godot game
Axe Ghost. On Steam!@axeghostgame.bsky.social

in response to my query godot updated the contact page to make clear that even for security bugs, ai-generated reports are not welcome. since all my reports are created that way, i won't be attempting coordinated disclosure with them any longer. more to follow.

large language model in the sense that i'm a large language animal

JCorvinus@jcorvinus.bsky.social

The name 'large language model' is so incredibly misleading. It's a neural sequence predictor that can learn to *use* language, but that's not what it *is*. Also people thinking 'stochastic probability' means 'wrong' or 'guess', when actually probability+iteration=convergence.

Axe Ghost. On Steam! reposted

My trad wife setting up the lighting for the chicken coop. She ties her bonnet on. She ties bonnets on all the hens. Me and the rooster applying our beard oil. Me and the rooster drinking black coffee

Axe Ghost. On Steam! reposted

Replying to @dame.is

Those who can completely “opt out” of or be least affected by societal change of this scale are those w/ privilege. This isn’t a condemnation, but it’s to say many militant anti-AI leaders are fostering an environment that will result in *less* privileged people being more negatively impacted by AI.

Axe Ghost. On Steam! reposted

Please, Cybersecurity Awareness Month Committee, you win. Make it stop. We’ve never been more aware.

jviide.iki.fi@jviide.iki.fi

October is Cybersecurity Awareness Month, and I think it's working!

yes. the rule *is* well tailored, coincidentally or not, to enable godot to say "this project doesn't rely on ai" with plausible deniability. and this might be a strategic consideration that's real for them, given their user base.

Cody@cody.chaos.net.nz

A rule that reports must be clear, concise, and with clear steps to allow verification would be reasonable. But current AI tools are more than capable of hitting that bar if people just try. The project's rules are not well tailored to achieve any reasonable aims.

here's the thing: godot's requirement of human written communications is supposed to work as a filter for 'does contributor understand what they're submitting' and I don't! not in the line-level way they intend at least. I have no interest in understanding that.

Axe Ghost. On Steam! reposted

how can we better communicate *why* an llm checking its own work, works? to fair-minded skeptics i think the intuition is: if t same system looks at it twice, the errors are going to be correlated! is the blocking part that the success of 'ok be a different kind of guy now' seems implausible?

Axe Ghost. On Steam!@axeghostgame.bsky.social

how i feel explaining llm -> llm adversarial review to non ai types

I suppose the issue, is non AI users can't understand how a person may simultaneously not understand/be interested in line-by-line code that touches the bug, & *still* have justified certainty that a finding is real (repro, antagonistic review, general llm-knowing) I claim to be in this position

Sqrt-1@sqrt-1.bsky.social

I am pro AI but this seems perfectly reasonable; they aren't banning AI submitted vulns at all, you just have to write them without AI. Prevents spam and ensures the submitter actually understands the vuln instead of off-loading review onto someone else.

in response to my query godot updated the contact page to make clear that even for security bugs, ai-generated reports are not welcome. since all my reports are created that way, i won't be attempting coordinated disclosure with them any longer. more to follow.

Axe Ghost. On Steam!@axeghostgame.bsky.social

if your OSS project doesn't accept AI-assisted vuln reports pls explicitly say so *ALONG WITH* an explicit statement that vulns found with AI assistance are released from any expectation of coordinated-disclosure embargo - the reporter may disclose whenever, at their own discretion.

pick one

Zach Weinersmith@zachweinersmith.bsky.social

a) that award committees aren't so smart b) that award committees don't care about quality c) that ai can write good fiction (honestly not my impression, but I've never tried to convince one to write a book) d) that it's all relative, and a crapshoot, and ai is just more crapshoot entries

Axe Ghost. On Steam! reposted

Replying to @davidcrespo.bsky.social

here is a demo of my "ah" tool showing: - terminal-synced dashboard with live HTML task tracker - side by side markdown editor - diff review UI - arbitrary HTML illustration all of which have integrated buttons/chat/comments that talk back to the running TUI agent. will make repo public soon

youtube.comah demoYouTube video by David Crespo

maybe this is already that!, but now I'm thinking of a 'chat interface' that's really a long html page w wherever arbitrary widgets/illustrations the agent deems helpful to add on its turn. with your regular text input fields inbetween.

conputer dipshit@davidcrespo.bsky.social

this is where Patchwork comes in. the tool I built is aimed at working alone but it doesn't have to be. imagine you're on a call, working on shared design documents (not text but arbitrary HTML with rich, interactive illustrations), all talking to the same agents prototyping for you in real time

Axe Ghost. On Steam! reposted

dame@dame.is

some of you actually do need an annoying guy popping out of a cartoon well reminding you of the nature of your existence, and you’re not in fact a country peasant carrying a bundle of sticks merely asking society to improve somewhat

Axe Ghost. On Steam! reposted

Do you have any photos in your camera roll that you expect would be difficult to geolocate? We’re looking for photo submissions for the Bellingcat December Geolocation Challenges - submit your images here: docs.google.com/forms/d/e/1F...

yes. 7 prototypes here. a maximum of three have a chance at being worked out into completed games, the rest aren't good enough - i wouldn't want to be represented by that work.

I'm currently sitting on prototypes for five different games. I could just release them as is. But I don't want to release anything I don't think is truly great, and AI can't speed up that process. It takes a lot of iteration and thought. So it might be years before all of them see the light of day.

1. creating a privately owned product from information in a commons is not, by itself, equivalent to enclosing that commons - the source material is still available 2. it has not been shown that externalities from the products (spam) have exceeded benefits (high quality submissions)

Peter K. G. Williams@pkgw.bsky.social

Likely the most important thing I've ever written — if you're an academic and the AI mania unsettles you, here's my best shot at presenting some intellectual tools that might help you articulate why. arxiv.org/abs/2610.27251

Axe Ghost. On Steam! reposted

A lot of people seem confused because AI both lowers the floor AND raises the ceiling. They se slop and assume thats all that's possible. But special is possible but takes effort! Don't listen to people who cannot imagine using a powerful tool to do anything but put minimal effort in.

This post is unavailable.

"A tech that lets a Zimbabwean teenager compete with a 40 something American suburbanite knowledge worker is going to test a lot of people's progressivism."

I don't think it's a coincidence that the further down the privilege ladder you go from American middle class the attitude towards AI improves. A tech that lets a Zimbabwean teenager compete with a 40 something American suburbanite knowledge worker is going to test a lot of people's progressivism.

On the Dangers of Stochastic Parrots:Can Language Models Be Too Big? Bender, McMillan-Major, Gebru, Shmitchell emphasis added: "If a large LM ... can manipulate linguistic form well enough to *cheat* its way through tests meant to require language understanding"

"what exactly BERT, GPT-3 and their kin are *learning* about linguistic structure" "large LMs’ ability to *learn* patterns of forms that humans associate with various biases and other harmful attitudes" "the risks of LMs *memorizing* training data"

Axe Ghost. On Steam! reposted

i wanna hammer on this too: most of everything you touch exists because industrialization made it cheap; your couch is factory-slop, as any carpenter will gladly tell you. your phone as well if you call me stupid, you'll use industrial slop to do so, even aside from 'bsky uses llms'

jae (hallowed)@fubarchitect.com

"write it yourself" i don't want it that bad! it's worth having specifically because the idea came to me at a moment when i can pay a gpu-haver to summon it via text elemental. it's not "worth it" in-itself like most software you use, which was largely made for money, not intrinsic value