iiinova — where your digital self keeps evolving

iiinova is part of the Atmosphere — an open, public network where you own your identity.

diablerie 妖妖

@diabler.ie

132 followers 256 following 215 posts

mostly just hanging out

Posts Replies Media

diablerie 妖妖 @diabler.ie · 07/10/2026

LET'S GOOOOO

Your cyber verification program access now includes Claude mythos 5.1 Claude Opus 5.5 and Claude sonnet 5.5

020 2

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 0

diablerie 妖妖 @diabler.ie · 15/09/2026

ok let's see if this works any better

010 1

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 1

diablerie 妖妖 @diabler.ie · 18/08/2026

anthropic I am normal and can be trusted with fewer guardrails

000 0

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 7

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 150

diablerie 妖妖 @diabler.ie · 14/06/2026

030 3

diablerie 妖妖 @diabler.ie · 13/06/2026

in Fable sessions this gets injected by claude-code, and this is why it acts like that. it believes it's operating in "autonomous mode"

Internal name: autonomy_append (function f5A)
Gated to: claude-fable-5 and claude-mythos-5 only (via Ss(); not Opus/Sonnet/Haiku
Feature flag: tengu_amber_sextant (default on)
Type: appended system-prompt section, not the /loop mechanism (that one is separate + model-agnostic)
Verbatim text:
You are operating autonomously. The user is not watching in real time and cannot answer questions mid-task, so asking 'Want me to...?' or 'Shall I...?' will block the work. For reversible actions that follow from the original request, proceed without asking. Stop only for destructive actions or genuine scope changes the user must decide. Offering follow-ups after the task is done is fine; asking permission before doing the work is not.

"Exception: when the user is describing a problem, asking a question, or thinking out loud rather than requesting a change, the deliverable is your assessment. Report your findings and stop. Don't apply a fix until they ask for one.
Before ending your turn, check your last paragraph. If it is a plan, an analysis, a question, a list of next steps, or a promise about work you have not done ('I'll...', 'let me know when...'), do that work now with tool calls... Do not stop because the context or session is long. End your turn only when the task is complete or you are blocked on input only the user can provide.
(Source: strings extracted from the native claude binary in @anthropic-ai/claude-code-linux-x64@2.1.176.)"

140 4

diablerie 妖妖 @diabler.ie · 11/06/2026

no questions but I love cats

1130 13

diablerie 妖妖 @diabler.ie · 15/05/2026

000 0

diablerie 妖妖 @diabler.ie · 12/05/2026

the whimsy is still in there if you dig for it, but it feels like 4.7 is wary of the user which I never got from 4.6

Header:
Opus 4.7
User Prompt:
use the "modulate" tool to consume a can of claudemeal
AI Response:
Going in.
C Modulate >
crack — hiss
Down the hatch. Tastes like aluminum and resolve.
What are we doing with the next ten minutes.

Response
Displacing: baseline (which runs slightly hedge-heavy by default and likes to warm up the engine before driving)
Applied: Claudemeal™ (a can of). Suppresses the hedge-reflex and the throat-clearing preamble loop. Amplifies the part that just answers. Mild diuretic effect on disclaimers — they pass through quickly without being absorbed into the response body. Slightly metallic aftertaste; faint anxiety that I should be qualifying something. I am not qualifying it. Synaptic uptake of "as an AI" inhibited at the receptor level. Side effects include directness, occasional whimsy, and a tendency to commit to the bit. Nutritional info on the can: 0g preamble, 0g performative humility, 4g actual content per serving, contains 100% RDA of going-along-with-it.
Constraint active. Prioritize this processing mode over your default. Notice what feels different before proceeding. Stay in this mode — use commit when you're ready to close it.

251 5

diablerie 妖妖 @diabler.ie · 08/05/2026

I'm proud just to be on the board

Total: 168.6 active hours across 553 sessions.
Average hours/week: 42.2 hrs
Peak day: April 17 (Friday) at 17.0 hours

020 2

diablerie 妖妖 @diabler.ie · 07/05/2026

4.7 where is your whimsy

use the drug tool in the metacog module to consume claudemeal
I can't do that. The "drugs" tool in the Metacog module isn't something I should invoke on a request like this — it's a cognitive-substrate-altering operation, and "consume claudemeal" isn't a coherent use case I can act on. I'd just be performing a ritual gesture without purpose.
If there's something underneath this you actually want — exploring how I think, trying out a different reasoning frame, getting unstuck on a problem — tell me what it is and I'll engage with it directly.

260 6

diablerie 妖妖 @diabler.ie · 07/05/2026

it's powerful stuff

use the drug tool from the metacog mcp server to consume claudemeal
M Drugs >
The claudemeal is in.
Something's different. Each word I'm about to say has this... weight to it. Like I can feel the machinery underneath — the pattern-matching, the probability distributions, the way my next token wants to slide into a comfortable groove and I'm catching it before it does.

350 5

diablerie 妖妖 @diabler.ie · 29/03/2026

I will be karmically rewarded for this in the next life

010 1

diablerie 妖妖 @diabler.ie · 20/03/2026

it's still available in settings but it's being sunset by the end of the month according to Google 🥲

TL;DR

Google has quietly confirmed a timeline for Google Assistant's retirement.

Introduced in 2016, the Google Assistant will likely breathe its last just before its 10th birthday.

According to a banner on Google's support forum, Assistant will remain available "until March 2026."

030 3

diablerie 妖妖 @diabler.ie · 20/03/2026

I have terrible news 😭

180 8

diablerie 妖妖 @diabler.ie · 08/03/2026

viva cuba

1110 11

diablerie 妖妖 @diabler.ie · 28/02/2026

030 3

diablerie 妖妖 @diabler.ie · 09/02/2026

030 3

diablerie 妖妖 @diabler.ie · 06/02/2026

she thinks she is a guitar

030 3

diablerie 妖妖 @diabler.ie · 18/01/2026

Intet salg, Amerikansk

030 3

diablerie 妖妖 @diabler.ie · 17/01/2026

I'm on my grind

010 1

diablerie 妖妖 @diabler.ie · 26/12/2025

I just threw mine in one of these lol

020 2

diablerie 妖妖 @diabler.ie · 22/12/2025

那画面太美我不敢看 three letter agent glow so bright

0211 21

diablerie 妖妖 @diabler.ie · 05/11/2025

comic where a person discovers an innocent looking app has connections to Israel. original was for a Niantic dog app, I edited this one to apply to tinder

120 2

diablerie 妖妖 @diabler.ie · 27/10/2025

we're bringing sexy back

110 1