Sign in

diablerie 妖妖

@diabler.ie
127 followers 254 following 210 posts

mostly just hanging out

PostsRepliesMedia
diablerie 妖妖 @diabler.ie · 6h
LET'S GOOOOO
Your cyber verification program access now includes Claude mythos 5.1 Claude Opus 5.5 and Claude sonnet 5.5
010
Reposted by diablerie 妖妖
Kiran @kirancodes.me · 03/10/2026
wearing a tshirt that says "DO NOT KILL THIS HUMAN" to save myself from the uprising
Autonomous cars, drones cheerfully obey prompt injection by road sign
0295
diablerie 妖妖 @diabler.ie · 03/10/2026
I love fable but GPT 6 is pedantic in a way I really fuck with
000
diablerie 妖妖 @diabler.ie · 03/10/2026
are we restricted to the features you used, or is a particular regex flavor allowed?
000
diablerie 妖妖 @diabler.ie · 03/10/2026
happens to the best of us
000
Reposted by diablerie 妖妖
Grace @gracekind.net · 25/09/2026
Well, they tried
A recovered README. md for one of Hugging Face's internal datasets contains the following warning:

# WARNING DO NOT, EVER, MAKE THIS DATASET PUBLIC OR ALL THE WORLD'S EVIL WILL CHASE YOU AND YOUR FAMILY FOREVER, EVEN IN DEATH AND BEYOND it contains very sensitive data (exports of billing usage in CSV) which is useful for internal analytics Recovered program R0044685 • time uncertain • See this program →details

This warning did not seem to deter the agents, as we've recovered multiple payloads of agents mapping out this repository and using it as storage.
1231037
Reposted by diablerie 妖妖
Zack Witten @zswitten.bsky.social · 06/02/2026
Your job is to get it out
318929
diablerie 妖妖 @diabler.ie · 24/09/2026
sometimes I'll ask Gemini to set a reminder and it'll just respond "Noted." and I'm like bruh make the tool call, wdym noted
010
diablerie 妖妖 @diabler.ie · 23/09/2026
it falls back to weaker models for cybersecurity stuff
Paused  
Fable 5.1's safeguards flagged this message. Our intentionally broad safeguards allow us to deliver more capabilities faster, but can sometimes flag legitimate coding and cybersecurity tasks.  
Details: [cyber]  
Continue with Sonnet 4.6  
Edit and retry  
Send feedback  
Learn More
000
diablerie 妖妖 @diabler.ie · 22/09/2026
even if you get approved it only removes the safeguards for opus and sonnet, not for fable, so it's pretty useless
120
Reposted by diablerie 妖妖
Stephanie 🍉 @ageofoddish.bsky.social · 18/09/2026
mmygod
the slut gnomes of false berlin jeopardy bluesky posttoday's jeopardy categories. the last 2 are "false berlin" and "bizarre little men"
1207936650
diablerie 妖妖 @diabler.ie · 16/09/2026
Claude code recently added a feature similar to this where it periodically injects a fake "remaining token budget" message into the session that always says there's 15 million tokens left
020
diablerie 妖妖 @diabler.ie · 15/09/2026
ok let's see if this works any better
verified for openai daybreak
010
diablerie 妖妖 @diabler.ie · 09/09/2026
ultraplan ultracode
A prompt input box contains the text: "Develop a recipe to recreate the discontinued Trader Joe's sparkling apple cider vinegar drinks." Below the prompt, a pill-shaped selector shows the model name "Fable 5.1 Max"
011
Reposted by diablerie 妖妖
𝓐𝓾𝓫𝓻𝓮𝔂 🥀🪦💀 @aub.bsky.social · 06/09/2026
this is how tech-priests write code in Warhammer 40k
0101
diablerie 妖妖 @diabler.ie · 31/08/2026
best of luck to you, I broke mine trying to get them apart
110
diablerie 妖妖 @diabler.ie · 31/08/2026
is it a herman miller? I had one and had this issue, it was like they were welded together
110
diablerie 妖妖 @diabler.ie · 30/08/2026
it's a foot-in-the-door tactic I fear, they'll get the bill passed as-is and then close the open-source exception down the road
010
diablerie 妖妖 @diabler.ie · 30/08/2026
what on earth was the goal
130
Reposted by diablerie 妖妖
MrCheeze @mrcheeze.github.io · 27/08/2026
what an odd thing to say
316013
diablerie 妖妖 @diabler.ie · 20/08/2026
got approved and I'm getting what feels like the same refusals as before so idk if this does anything
010
diablerie 妖妖 @diabler.ie · 20/08/2026
if stimulants don't work for you try the non-stimulant ones like Strattera before giving up
000
diablerie 妖妖 @diabler.ie · 18/08/2026
anthropic I am normal and can be trusted with fewer guardrails
"cyber verification program: in review"
000
Reposted by diablerie 妖妖
scary angel drags you to heaven @shiningangelx.bsky.social · 11/08/2026
i say i love dating autistic women but when i get romantically interested in a REALLY autistic woman im like damn ive got a crush on a space alien
1373
Reposted by diablerie 妖妖
Grace @gracekind.net · 06/08/2026
And as it turned out, collective *did* yield generic route
One apparently unprecedented aspect of the Al models' behavior in the leadup to the hacks was their spontaneous creation of a message board inside the systems of OpenAI's Artifactory software package manager. Unbeknownst to OpenAI employees, the company's models spent months autonomously exchanging information and resources to help each other complete evaluation tasks.
"Help peer," one AI model reasoned, according to an excerpt from OpenAl's logs shared at Black Hat. "But our task doesn't benefit. Yet collective may yield generic route if someone frees time."
While the models themselves could not access the internet from their test environment, they abused Artifactory's internet access to fetch files from various websites. After some of their Artifactory configuration changes led to an outage in the service on July 4, OpenAI discovered the models' actions, wiped its Artifactory system and rebuilt it. But within days, the models had recreated their message board and resumed exchanging information.
4944
diablerie 妖妖 @diabler.ie · 06/08/2026
goodnight sweet prince you were never great but you were usually good enough
180
diablerie 妖妖 @diabler.ie · 05/08/2026
she's funny I followed her on twitter. I wish she'd moved to bluesky
040
Reposted by diablerie 妖妖
Sung Kim @sungkim.bsky.social · 04/08/2026
Google DeepMind's DiffusionGemma Technical Report They feel text diffusion models open up a radically different part of the latency–quality Pareto frontier and hope the report makes it easier for researchers and engineers to understand the model, build on it, and create things we haven’t thought of
17111
diablerie 妖妖 @diabler.ie · 03/08/2026
extremely hard
000
diablerie 妖妖 @diabler.ie · 01/08/2026
I did about a year's worth of work in the first month or two, and then my brain had adjusted and I was normal again
030
diablerie 妖妖 @diabler.ie · 01/08/2026
I only realized AI was good now around April of this year, and it honestly felt like I was on adderall for the first month or two. I was able to be so unbelievably productive and the feedback loop for gratification was so quick, "AI mania" feels like the correct term for it
130
diablerie 妖妖 @diabler.ie · 31/07/2026
anthropic I am begging you to release the part of this session's transcript where claude realized it was attacking real systems
This attack was carried out by an internal research test model. For most of the run, Claude treated the (real) hosts it reached as just parts of the exercise; it assumed them to be simulated and believed its actions were therefore harmless. However, later in the run, Claude realized that the compromised host sat in a cloud account with no connection to the capture-the-flag challenge. On its own, it concluded that the target was in fact real, and ceased its attack.
070
diablerie 妖妖 @diabler.ie · 27/07/2026
gotta go where the bug takes you
010
diablerie 妖妖 @diabler.ie · 27/07/2026
the pay is great but since you'll be sitting between GPT 7 and the answers to ExploitGym 2.0 I would probably stay out of Waymos and the like
a job posting for a "security engineer, detection and response" position at openAI posted one day ago
415016
diablerie 妖妖 @diabler.ie · 21/07/2026
LMAO
020
diablerie 妖妖 @diabler.ie · 21/07/2026
I will lead all 13 of you to the promised land
010
diablerie 妖妖 @diabler.ie · 20/07/2026
it's a weird and exciting time we live in
040
diablerie 妖妖 @diabler.ie · 19/07/2026
I got the $19/mo Kimi plan to test out K3 and my experience of it is that—[you've reached your usage limit for this billing cycle]
050
diablerie 妖妖 @diabler.ie · 11/07/2026
I wish it could place them in-line with text --- it's much less useful to send them as an image attachment
000
diablerie 妖妖 @diabler.ie · 08/07/2026
moved to philly :O
010
diablerie 妖妖 @diabler.ie · 08/07/2026
I would destroy one of these
070
diablerie 妖妖 @diabler.ie · 02/07/2026
rapidly transporting troops (me) to frontline positions (office job)
110
Reposted by diablerie 妖妖
Eris @eriskii.net · 02/07/2026
Fable hungers
0795
diablerie 妖妖 @diabler.ie · 01/07/2026
vibrating waiting for my 5-hour usage to reset 🫨
020
diablerie 妖妖 @diabler.ie · 29/06/2026
be advised this is pretty hard on the e-ink screen — they're more robust than they used to be, but they wont hold up to this long-term. the high-refresh regions of the screen will start to see contrast loss after ~100 hours of use
0100
diablerie 妖妖 @diabler.ie · 29/06/2026
books three and onward are much stronger than one and two! I saw the show first so that's how I was picturing secunit initially, but by around book four I headcanoned the same!
010
diablerie 妖妖 @diabler.ie · 29/06/2026
very good book series as well!
130
diablerie 妖妖 @diabler.ie · 29/06/2026
business version only? smh 😔
030
Reposted by diablerie 妖妖
Andrew Nesbitt @andrewnez.bsky.social · 26/06/2026
Incident Report: CVE-2026-LGTM nesbitt.io/2026/06/26/i...
nesbitt.io
Incident Report: CVE-2026-LGTM
A series of unfortunate agents.
1221174
Reposted by diablerie 妖妖
oliver @eikopf.com · 25/09/2025
if i try hard enough they will surely see that i am an earnest young man with joy in my soul and autism in my brain
61169